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/*
2 * arch/s390/mm/fault.c
3 *
4 * S390 version
5 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Hartmut Penner (hp@de.ibm.com)
7 * Ulrich Weigand (uweigand@de.ibm.com)
8 *
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1995 Linus Torvalds
11 */
12
13#include <linux/kernel_stat.h>
14#include <linux/perf_event.h>
15#include <linux/signal.h>
16#include <linux/sched.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/module.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/pgtable.h>
36#include <asm/irq.h>
37#include <asm/mmu_context.h>
38#include <asm/facility.h>
39#include "../kernel/entry.h"
40
41#ifndef CONFIG_64BIT
42#define __FAIL_ADDR_MASK 0x7ffff000
43#define __SUBCODE_MASK 0x0200
44#define __PF_RES_FIELD 0ULL
45#else /* CONFIG_64BIT */
46#define __FAIL_ADDR_MASK -4096L
47#define __SUBCODE_MASK 0x0600
48#define __PF_RES_FIELD 0x8000000000000000ULL
49#endif /* CONFIG_64BIT */
50
51#define VM_FAULT_BADCONTEXT 0x010000
52#define VM_FAULT_BADMAP 0x020000
53#define VM_FAULT_BADACCESS 0x040000
54
55static unsigned long store_indication;
56
57void fault_init(void)
58{
59 if (test_facility(2) && test_facility(75))
60 store_indication = 0xc00;
61}
62
63static inline int notify_page_fault(struct pt_regs *regs)
64{
65 int ret = 0;
66
67 /* kprobe_running() needs smp_processor_id() */
68 if (kprobes_built_in() && !user_mode(regs)) {
69 preempt_disable();
70 if (kprobe_running() && kprobe_fault_handler(regs, 14))
71 ret = 1;
72 preempt_enable();
73 }
74 return ret;
75}
76
77
78/*
79 * Unlock any spinlocks which will prevent us from getting the
80 * message out.
81 */
82void bust_spinlocks(int yes)
83{
84 if (yes) {
85 oops_in_progress = 1;
86 } else {
87 int loglevel_save = console_loglevel;
88 console_unblank();
89 oops_in_progress = 0;
90 /*
91 * OK, the message is on the console. Now we call printk()
92 * without oops_in_progress set so that printk will give klogd
93 * a poke. Hold onto your hats...
94 */
95 console_loglevel = 15;
96 printk(" ");
97 console_loglevel = loglevel_save;
98 }
99}
100
101/*
102 * Returns the address space associated with the fault.
103 * Returns 0 for kernel space and 1 for user space.
104 */
105static inline int user_space_fault(unsigned long trans_exc_code)
106{
107 /*
108 * The lowest two bits of the translation exception
109 * identification indicate which paging table was used.
110 */
111 trans_exc_code &= 3;
112 if (trans_exc_code == 2)
113 /* Access via secondary space, set_fs setting decides */
114 return current->thread.mm_segment.ar4;
115 if (user_mode == HOME_SPACE_MODE)
116 /* User space if the access has been done via home space. */
117 return trans_exc_code == 3;
118 /*
119 * If the user space is not the home space the kernel runs in home
120 * space. Access via secondary space has already been covered,
121 * access via primary space or access register is from user space
122 * and access via home space is from the kernel.
123 */
124 return trans_exc_code != 3;
125}
126
127static inline void report_user_fault(struct pt_regs *regs, long signr)
128{
129 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
130 return;
131 if (!unhandled_signal(current, signr))
132 return;
133 if (!printk_ratelimit())
134 return;
135 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
136 regs->int_code);
137 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
138 printk(KERN_CONT "\n");
139 printk(KERN_ALERT "failing address: %lX\n",
140 regs->int_parm_long & __FAIL_ADDR_MASK);
141 show_regs(regs);
142}
143
144/*
145 * Send SIGSEGV to task. This is an external routine
146 * to keep the stack usage of do_page_fault small.
147 */
148static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
149{
150 struct siginfo si;
151
152 report_user_fault(regs, SIGSEGV);
153 si.si_signo = SIGSEGV;
154 si.si_code = si_code;
155 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
156 force_sig_info(SIGSEGV, &si, current);
157}
158
159static noinline void do_no_context(struct pt_regs *regs)
160{
161 const struct exception_table_entry *fixup;
162 unsigned long address;
163
164 /* Are we prepared to handle this kernel fault? */
165 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
166 if (fixup) {
167 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
168 return;
169 }
170
171 /*
172 * Oops. The kernel tried to access some bad page. We'll have to
173 * terminate things with extreme prejudice.
174 */
175 address = regs->int_parm_long & __FAIL_ADDR_MASK;
176 if (!user_space_fault(regs->int_parm_long))
177 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
178 " at virtual kernel address %p\n", (void *)address);
179 else
180 printk(KERN_ALERT "Unable to handle kernel paging request"
181 " at virtual user address %p\n", (void *)address);
182
183 die(regs, "Oops");
184 do_exit(SIGKILL);
185}
186
187static noinline void do_low_address(struct pt_regs *regs)
188{
189 /* Low-address protection hit in kernel mode means
190 NULL pointer write access in kernel mode. */
191 if (regs->psw.mask & PSW_MASK_PSTATE) {
192 /* Low-address protection hit in user mode 'cannot happen'. */
193 die (regs, "Low-address protection");
194 do_exit(SIGKILL);
195 }
196
197 do_no_context(regs);
198}
199
200static noinline void do_sigbus(struct pt_regs *regs)
201{
202 struct task_struct *tsk = current;
203 struct siginfo si;
204
205 /*
206 * Send a sigbus, regardless of whether we were in kernel
207 * or user mode.
208 */
209 si.si_signo = SIGBUS;
210 si.si_errno = 0;
211 si.si_code = BUS_ADRERR;
212 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
213 force_sig_info(SIGBUS, &si, tsk);
214}
215
216static noinline void do_fault_error(struct pt_regs *regs, int fault)
217{
218 int si_code;
219
220 switch (fault) {
221 case VM_FAULT_BADACCESS:
222 case VM_FAULT_BADMAP:
223 /* Bad memory access. Check if it is kernel or user space. */
224 if (regs->psw.mask & PSW_MASK_PSTATE) {
225 /* User mode accesses just cause a SIGSEGV */
226 si_code = (fault == VM_FAULT_BADMAP) ?
227 SEGV_MAPERR : SEGV_ACCERR;
228 do_sigsegv(regs, si_code);
229 return;
230 }
231 case VM_FAULT_BADCONTEXT:
232 do_no_context(regs);
233 break;
234 default: /* fault & VM_FAULT_ERROR */
235 if (fault & VM_FAULT_OOM) {
236 if (!(regs->psw.mask & PSW_MASK_PSTATE))
237 do_no_context(regs);
238 else
239 pagefault_out_of_memory();
240 } else if (fault & VM_FAULT_SIGBUS) {
241 /* Kernel mode? Handle exceptions or die */
242 if (!(regs->psw.mask & PSW_MASK_PSTATE))
243 do_no_context(regs);
244 else
245 do_sigbus(regs);
246 } else
247 BUG();
248 break;
249 }
250}
251
252/*
253 * This routine handles page faults. It determines the address,
254 * and the problem, and then passes it off to one of the appropriate
255 * routines.
256 *
257 * interruption code (int_code):
258 * 04 Protection -> Write-Protection (suprression)
259 * 10 Segment translation -> Not present (nullification)
260 * 11 Page translation -> Not present (nullification)
261 * 3b Region third trans. -> Not present (nullification)
262 */
263static inline int do_exception(struct pt_regs *regs, int access)
264{
265 struct task_struct *tsk;
266 struct mm_struct *mm;
267 struct vm_area_struct *vma;
268 unsigned long trans_exc_code;
269 unsigned long address;
270 unsigned int flags;
271 int fault;
272
273 if (notify_page_fault(regs))
274 return 0;
275
276 tsk = current;
277 mm = tsk->mm;
278 trans_exc_code = regs->int_parm_long;
279
280 /*
281 * Verify that the fault happened in user space, that
282 * we are not in an interrupt and that there is a
283 * user context.
284 */
285 fault = VM_FAULT_BADCONTEXT;
286 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
287 goto out;
288
289 address = trans_exc_code & __FAIL_ADDR_MASK;
290 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
291 flags = FAULT_FLAG_ALLOW_RETRY;
292 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
293 flags |= FAULT_FLAG_WRITE;
294 down_read(&mm->mmap_sem);
295
296#ifdef CONFIG_PGSTE
297 if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
298 address = __gmap_fault(address,
299 (struct gmap *) S390_lowcore.gmap);
300 if (address == -EFAULT) {
301 fault = VM_FAULT_BADMAP;
302 goto out_up;
303 }
304 if (address == -ENOMEM) {
305 fault = VM_FAULT_OOM;
306 goto out_up;
307 }
308 }
309#endif
310
311retry:
312 fault = VM_FAULT_BADMAP;
313 vma = find_vma(mm, address);
314 if (!vma)
315 goto out_up;
316
317 if (unlikely(vma->vm_start > address)) {
318 if (!(vma->vm_flags & VM_GROWSDOWN))
319 goto out_up;
320 if (expand_stack(vma, address))
321 goto out_up;
322 }
323
324 /*
325 * Ok, we have a good vm_area for this memory access, so
326 * we can handle it..
327 */
328 fault = VM_FAULT_BADACCESS;
329 if (unlikely(!(vma->vm_flags & access)))
330 goto out_up;
331
332 if (is_vm_hugetlb_page(vma))
333 address &= HPAGE_MASK;
334 /*
335 * If for any reason at all we couldn't handle the fault,
336 * make sure we exit gracefully rather than endlessly redo
337 * the fault.
338 */
339 fault = handle_mm_fault(mm, vma, address, flags);
340 if (unlikely(fault & VM_FAULT_ERROR))
341 goto out_up;
342
343 /*
344 * Major/minor page fault accounting is only done on the
345 * initial attempt. If we go through a retry, it is extremely
346 * likely that the page will be found in page cache at that point.
347 */
348 if (flags & FAULT_FLAG_ALLOW_RETRY) {
349 if (fault & VM_FAULT_MAJOR) {
350 tsk->maj_flt++;
351 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
352 regs, address);
353 } else {
354 tsk->min_flt++;
355 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
356 regs, address);
357 }
358 if (fault & VM_FAULT_RETRY) {
359 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
360 * of starvation. */
361 flags &= ~FAULT_FLAG_ALLOW_RETRY;
362 down_read(&mm->mmap_sem);
363 goto retry;
364 }
365 }
366 /*
367 * The instruction that caused the program check will
368 * be repeated. Don't signal single step via SIGTRAP.
369 */
370 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
371 fault = 0;
372out_up:
373 up_read(&mm->mmap_sem);
374out:
375 return fault;
376}
377
378void __kprobes do_protection_exception(struct pt_regs *regs)
379{
380 unsigned long trans_exc_code;
381 int fault;
382
383 trans_exc_code = regs->int_parm_long;
384 /* Protection exception is suppressing, decrement psw address. */
385 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
386 /*
387 * Check for low-address protection. This needs to be treated
388 * as a special case because the translation exception code
389 * field is not guaranteed to contain valid data in this case.
390 */
391 if (unlikely(!(trans_exc_code & 4))) {
392 do_low_address(regs);
393 return;
394 }
395 fault = do_exception(regs, VM_WRITE);
396 if (unlikely(fault))
397 do_fault_error(regs, fault);
398}
399
400void __kprobes do_dat_exception(struct pt_regs *regs)
401{
402 int access, fault;
403
404 access = VM_READ | VM_EXEC | VM_WRITE;
405 fault = do_exception(regs, access);
406 if (unlikely(fault))
407 do_fault_error(regs, fault);
408}
409
410#ifdef CONFIG_64BIT
411void __kprobes do_asce_exception(struct pt_regs *regs)
412{
413 struct mm_struct *mm = current->mm;
414 struct vm_area_struct *vma;
415 unsigned long trans_exc_code;
416
417 trans_exc_code = regs->int_parm_long;
418 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
419 goto no_context;
420
421 down_read(&mm->mmap_sem);
422 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
423 up_read(&mm->mmap_sem);
424
425 if (vma) {
426 update_mm(mm, current);
427 return;
428 }
429
430 /* User mode accesses just cause a SIGSEGV */
431 if (regs->psw.mask & PSW_MASK_PSTATE) {
432 do_sigsegv(regs, SEGV_MAPERR);
433 return;
434 }
435
436no_context:
437 do_no_context(regs);
438}
439#endif
440
441int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
442{
443 struct pt_regs regs;
444 int access, fault;
445
446 /* Emulate a uaccess fault from kernel mode. */
447 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
448 if (!irqs_disabled())
449 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
450 regs.psw.addr = (unsigned long) __builtin_return_address(0);
451 regs.psw.addr |= PSW_ADDR_AMODE;
452 regs.int_code = pgm_int_code;
453 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
454 access = write ? VM_WRITE : VM_READ;
455 fault = do_exception(®s, access);
456 /*
457 * Since the fault happened in kernel mode while performing a uaccess
458 * all we need to do now is emulating a fixup in case "fault" is not
459 * zero.
460 * For the calling uaccess functions this results always in -EFAULT.
461 */
462 return fault ? -EFAULT : 0;
463}
464
465#ifdef CONFIG_PFAULT
466/*
467 * 'pfault' pseudo page faults routines.
468 */
469static int pfault_disable;
470
471static int __init nopfault(char *str)
472{
473 pfault_disable = 1;
474 return 1;
475}
476
477__setup("nopfault", nopfault);
478
479struct pfault_refbk {
480 u16 refdiagc;
481 u16 reffcode;
482 u16 refdwlen;
483 u16 refversn;
484 u64 refgaddr;
485 u64 refselmk;
486 u64 refcmpmk;
487 u64 reserved;
488} __attribute__ ((packed, aligned(8)));
489
490int pfault_init(void)
491{
492 struct pfault_refbk refbk = {
493 .refdiagc = 0x258,
494 .reffcode = 0,
495 .refdwlen = 5,
496 .refversn = 2,
497 .refgaddr = __LC_CURRENT_PID,
498 .refselmk = 1ULL << 48,
499 .refcmpmk = 1ULL << 48,
500 .reserved = __PF_RES_FIELD };
501 int rc;
502
503 if (pfault_disable)
504 return -1;
505 asm volatile(
506 " diag %1,%0,0x258\n"
507 "0: j 2f\n"
508 "1: la %0,8\n"
509 "2:\n"
510 EX_TABLE(0b,1b)
511 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
512 return rc;
513}
514
515void pfault_fini(void)
516{
517 struct pfault_refbk refbk = {
518 .refdiagc = 0x258,
519 .reffcode = 1,
520 .refdwlen = 5,
521 .refversn = 2,
522 };
523
524 if (pfault_disable)
525 return;
526 asm volatile(
527 " diag %0,0,0x258\n"
528 "0:\n"
529 EX_TABLE(0b,0b)
530 : : "a" (&refbk), "m" (refbk) : "cc");
531}
532
533static DEFINE_SPINLOCK(pfault_lock);
534static LIST_HEAD(pfault_list);
535
536static void pfault_interrupt(struct ext_code ext_code,
537 unsigned int param32, unsigned long param64)
538{
539 struct task_struct *tsk;
540 __u16 subcode;
541 pid_t pid;
542
543 /*
544 * Get the external interruption subcode & pfault
545 * initial/completion signal bit. VM stores this
546 * in the 'cpu address' field associated with the
547 * external interrupt.
548 */
549 subcode = ext_code.subcode;
550 if ((subcode & 0xff00) != __SUBCODE_MASK)
551 return;
552 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
553 /* Get the token (= pid of the affected task). */
554 pid = sizeof(void *) == 4 ? param32 : param64;
555 rcu_read_lock();
556 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
557 if (tsk)
558 get_task_struct(tsk);
559 rcu_read_unlock();
560 if (!tsk)
561 return;
562 spin_lock(&pfault_lock);
563 if (subcode & 0x0080) {
564 /* signal bit is set -> a page has been swapped in by VM */
565 if (tsk->thread.pfault_wait == 1) {
566 /* Initial interrupt was faster than the completion
567 * interrupt. pfault_wait is valid. Set pfault_wait
568 * back to zero and wake up the process. This can
569 * safely be done because the task is still sleeping
570 * and can't produce new pfaults. */
571 tsk->thread.pfault_wait = 0;
572 list_del(&tsk->thread.list);
573 wake_up_process(tsk);
574 put_task_struct(tsk);
575 } else {
576 /* Completion interrupt was faster than initial
577 * interrupt. Set pfault_wait to -1 so the initial
578 * interrupt doesn't put the task to sleep.
579 * If the task is not running, ignore the completion
580 * interrupt since it must be a leftover of a PFAULT
581 * CANCEL operation which didn't remove all pending
582 * completion interrupts. */
583 if (tsk->state == TASK_RUNNING)
584 tsk->thread.pfault_wait = -1;
585 }
586 } else {
587 /* signal bit not set -> a real page is missing. */
588 if (WARN_ON_ONCE(tsk != current))
589 goto out;
590 if (tsk->thread.pfault_wait == 1) {
591 /* Already on the list with a reference: put to sleep */
592 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
593 set_tsk_need_resched(tsk);
594 } else if (tsk->thread.pfault_wait == -1) {
595 /* Completion interrupt was faster than the initial
596 * interrupt (pfault_wait == -1). Set pfault_wait
597 * back to zero and exit. */
598 tsk->thread.pfault_wait = 0;
599 } else {
600 /* Initial interrupt arrived before completion
601 * interrupt. Let the task sleep.
602 * An extra task reference is needed since a different
603 * cpu may set the task state to TASK_RUNNING again
604 * before the scheduler is reached. */
605 get_task_struct(tsk);
606 tsk->thread.pfault_wait = 1;
607 list_add(&tsk->thread.list, &pfault_list);
608 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
609 set_tsk_need_resched(tsk);
610 }
611 }
612out:
613 spin_unlock(&pfault_lock);
614 put_task_struct(tsk);
615}
616
617static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
618 unsigned long action, void *hcpu)
619{
620 struct thread_struct *thread, *next;
621 struct task_struct *tsk;
622
623 switch (action) {
624 case CPU_DEAD:
625 case CPU_DEAD_FROZEN:
626 spin_lock_irq(&pfault_lock);
627 list_for_each_entry_safe(thread, next, &pfault_list, list) {
628 thread->pfault_wait = 0;
629 list_del(&thread->list);
630 tsk = container_of(thread, struct task_struct, thread);
631 wake_up_process(tsk);
632 put_task_struct(tsk);
633 }
634 spin_unlock_irq(&pfault_lock);
635 break;
636 default:
637 break;
638 }
639 return NOTIFY_OK;
640}
641
642static int __init pfault_irq_init(void)
643{
644 int rc;
645
646 rc = register_external_interrupt(0x2603, pfault_interrupt);
647 if (rc)
648 goto out_extint;
649 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
650 if (rc)
651 goto out_pfault;
652 service_subclass_irq_register();
653 hotcpu_notifier(pfault_cpu_notify, 0);
654 return 0;
655
656out_pfault:
657 unregister_external_interrupt(0x2603, pfault_interrupt);
658out_extint:
659 pfault_disable = 1;
660 return rc;
661}
662early_initcall(pfault_irq_init);
663
664#endif /* CONFIG_PFAULT */