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