Loading...
1/*
2 * S390 version
3 * Copyright IBM Corp. 1999
4 * Author(s): Hartmut Penner (hp@de.ibm.com)
5 * Ulrich Weigand (uweigand@de.ibm.com)
6 *
7 * Derived from "arch/i386/mm/fault.c"
8 * Copyright (C) 1995 Linus Torvalds
9 */
10
11#include <linux/kernel_stat.h>
12#include <linux/perf_event.h>
13#include <linux/signal.h>
14#include <linux/sched.h>
15#include <linux/kernel.h>
16#include <linux/errno.h>
17#include <linux/string.h>
18#include <linux/types.h>
19#include <linux/ptrace.h>
20#include <linux/mman.h>
21#include <linux/mm.h>
22#include <linux/compat.h>
23#include <linux/smp.h>
24#include <linux/kdebug.h>
25#include <linux/init.h>
26#include <linux/console.h>
27#include <linux/module.h>
28#include <linux/hardirq.h>
29#include <linux/kprobes.h>
30#include <linux/uaccess.h>
31#include <linux/hugetlb.h>
32#include <asm/asm-offsets.h>
33#include <asm/diag.h>
34#include <asm/pgtable.h>
35#include <asm/gmap.h>
36#include <asm/irq.h>
37#include <asm/mmu_context.h>
38#include <asm/facility.h>
39#include "../kernel/entry.h"
40
41#define __FAIL_ADDR_MASK -4096L
42#define __SUBCODE_MASK 0x0600
43#define __PF_RES_FIELD 0x8000000000000000ULL
44
45#define VM_FAULT_BADCONTEXT 0x010000
46#define VM_FAULT_BADMAP 0x020000
47#define VM_FAULT_BADACCESS 0x040000
48#define VM_FAULT_SIGNAL 0x080000
49#define VM_FAULT_PFAULT 0x100000
50
51static unsigned long store_indication __read_mostly;
52
53static int __init fault_init(void)
54{
55 if (test_facility(75))
56 store_indication = 0xc00;
57 return 0;
58}
59early_initcall(fault_init);
60
61static inline int notify_page_fault(struct pt_regs *regs)
62{
63 int ret = 0;
64
65 /* kprobe_running() needs smp_processor_id() */
66 if (kprobes_built_in() && !user_mode(regs)) {
67 preempt_disable();
68 if (kprobe_running() && kprobe_fault_handler(regs, 14))
69 ret = 1;
70 preempt_enable();
71 }
72 return ret;
73}
74
75
76/*
77 * Unlock any spinlocks which will prevent us from getting the
78 * message out.
79 */
80void bust_spinlocks(int yes)
81{
82 if (yes) {
83 oops_in_progress = 1;
84 } else {
85 int loglevel_save = console_loglevel;
86 console_unblank();
87 oops_in_progress = 0;
88 /*
89 * OK, the message is on the console. Now we call printk()
90 * without oops_in_progress set so that printk will give klogd
91 * a poke. Hold onto your hats...
92 */
93 console_loglevel = 15;
94 printk(" ");
95 console_loglevel = loglevel_save;
96 }
97}
98
99/*
100 * Returns the address space associated with the fault.
101 * Returns 0 for kernel space and 1 for user space.
102 */
103static inline int user_space_fault(struct pt_regs *regs)
104{
105 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 = regs->int_parm_long & 3;
112 if (trans_exc_code == 3) /* home space -> kernel */
113 return 0;
114 if (user_mode(regs))
115 return 1;
116 if (trans_exc_code == 2) /* secondary space -> set_fs */
117 return current->thread.mm_segment.ar4;
118 if (current->flags & PF_VCPU)
119 return 1;
120 return 0;
121}
122
123static int bad_address(void *p)
124{
125 unsigned long dummy;
126
127 return probe_kernel_address((unsigned long *)p, dummy);
128}
129
130static void dump_pagetable(unsigned long asce, unsigned long address)
131{
132 unsigned long *table = __va(asce & PAGE_MASK);
133
134 pr_alert("AS:%016lx ", asce);
135 switch (asce & _ASCE_TYPE_MASK) {
136 case _ASCE_TYPE_REGION1:
137 table = table + ((address >> 53) & 0x7ff);
138 if (bad_address(table))
139 goto bad;
140 pr_cont("R1:%016lx ", *table);
141 if (*table & _REGION_ENTRY_INVALID)
142 goto out;
143 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
144 /* fallthrough */
145 case _ASCE_TYPE_REGION2:
146 table = table + ((address >> 42) & 0x7ff);
147 if (bad_address(table))
148 goto bad;
149 pr_cont("R2:%016lx ", *table);
150 if (*table & _REGION_ENTRY_INVALID)
151 goto out;
152 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
153 /* fallthrough */
154 case _ASCE_TYPE_REGION3:
155 table = table + ((address >> 31) & 0x7ff);
156 if (bad_address(table))
157 goto bad;
158 pr_cont("R3:%016lx ", *table);
159 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
160 goto out;
161 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
162 /* fallthrough */
163 case _ASCE_TYPE_SEGMENT:
164 table = table + ((address >> 20) & 0x7ff);
165 if (bad_address(table))
166 goto bad;
167 pr_cont("S:%016lx ", *table);
168 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
169 goto out;
170 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
171 }
172 table = table + ((address >> 12) & 0xff);
173 if (bad_address(table))
174 goto bad;
175 pr_cont("P:%016lx ", *table);
176out:
177 pr_cont("\n");
178 return;
179bad:
180 pr_cont("BAD\n");
181}
182
183static void dump_fault_info(struct pt_regs *regs)
184{
185 unsigned long asce;
186
187 pr_alert("Failing address: %016lx TEID: %016lx\n",
188 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
189 pr_alert("Fault in ");
190 switch (regs->int_parm_long & 3) {
191 case 3:
192 pr_cont("home space ");
193 break;
194 case 2:
195 pr_cont("secondary space ");
196 break;
197 case 1:
198 pr_cont("access register ");
199 break;
200 case 0:
201 pr_cont("primary space ");
202 break;
203 }
204 pr_cont("mode while using ");
205 if (!user_space_fault(regs)) {
206 asce = S390_lowcore.kernel_asce;
207 pr_cont("kernel ");
208 }
209#ifdef CONFIG_PGSTE
210 else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
211 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
212 asce = gmap->asce;
213 pr_cont("gmap ");
214 }
215#endif
216 else {
217 asce = S390_lowcore.user_asce;
218 pr_cont("user ");
219 }
220 pr_cont("ASCE.\n");
221 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
222}
223
224int show_unhandled_signals = 1;
225
226void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
227{
228 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
229 return;
230 if (!unhandled_signal(current, signr))
231 return;
232 if (!printk_ratelimit())
233 return;
234 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
235 regs->int_code & 0xffff, regs->int_code >> 17);
236 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
237 printk(KERN_CONT "\n");
238 if (is_mm_fault)
239 dump_fault_info(regs);
240 show_regs(regs);
241}
242
243/*
244 * Send SIGSEGV to task. This is an external routine
245 * to keep the stack usage of do_page_fault small.
246 */
247static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
248{
249 struct siginfo si;
250
251 report_user_fault(regs, SIGSEGV, 1);
252 si.si_signo = SIGSEGV;
253 si.si_code = si_code;
254 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
255 force_sig_info(SIGSEGV, &si, current);
256}
257
258static noinline void do_no_context(struct pt_regs *regs)
259{
260 const struct exception_table_entry *fixup;
261
262 /* Are we prepared to handle this kernel fault? */
263 fixup = search_exception_tables(regs->psw.addr);
264 if (fixup) {
265 regs->psw.addr = extable_fixup(fixup);
266 return;
267 }
268
269 /*
270 * Oops. The kernel tried to access some bad page. We'll have to
271 * terminate things with extreme prejudice.
272 */
273 if (!user_space_fault(regs))
274 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
275 " in virtual kernel address space\n");
276 else
277 printk(KERN_ALERT "Unable to handle kernel paging request"
278 " in virtual user address space\n");
279 dump_fault_info(regs);
280 die(regs, "Oops");
281 do_exit(SIGKILL);
282}
283
284static noinline void do_low_address(struct pt_regs *regs)
285{
286 /* Low-address protection hit in kernel mode means
287 NULL pointer write access in kernel mode. */
288 if (regs->psw.mask & PSW_MASK_PSTATE) {
289 /* Low-address protection hit in user mode 'cannot happen'. */
290 die (regs, "Low-address protection");
291 do_exit(SIGKILL);
292 }
293
294 do_no_context(regs);
295}
296
297static noinline void do_sigbus(struct pt_regs *regs)
298{
299 struct task_struct *tsk = current;
300 struct siginfo si;
301
302 /*
303 * Send a sigbus, regardless of whether we were in kernel
304 * or user mode.
305 */
306 si.si_signo = SIGBUS;
307 si.si_errno = 0;
308 si.si_code = BUS_ADRERR;
309 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
310 force_sig_info(SIGBUS, &si, tsk);
311}
312
313static noinline void do_fault_error(struct pt_regs *regs, int fault)
314{
315 int si_code;
316
317 switch (fault) {
318 case VM_FAULT_BADACCESS:
319 case VM_FAULT_BADMAP:
320 /* Bad memory access. Check if it is kernel or user space. */
321 if (user_mode(regs)) {
322 /* User mode accesses just cause a SIGSEGV */
323 si_code = (fault == VM_FAULT_BADMAP) ?
324 SEGV_MAPERR : SEGV_ACCERR;
325 do_sigsegv(regs, si_code);
326 return;
327 }
328 case VM_FAULT_BADCONTEXT:
329 case VM_FAULT_PFAULT:
330 do_no_context(regs);
331 break;
332 case VM_FAULT_SIGNAL:
333 if (!user_mode(regs))
334 do_no_context(regs);
335 break;
336 default: /* fault & VM_FAULT_ERROR */
337 if (fault & VM_FAULT_OOM) {
338 if (!user_mode(regs))
339 do_no_context(regs);
340 else
341 pagefault_out_of_memory();
342 } else if (fault & VM_FAULT_SIGSEGV) {
343 /* Kernel mode? Handle exceptions or die */
344 if (!user_mode(regs))
345 do_no_context(regs);
346 else
347 do_sigsegv(regs, SEGV_MAPERR);
348 } else if (fault & VM_FAULT_SIGBUS) {
349 /* Kernel mode? Handle exceptions or die */
350 if (!user_mode(regs))
351 do_no_context(regs);
352 else
353 do_sigbus(regs);
354 } else
355 BUG();
356 break;
357 }
358}
359
360/*
361 * This routine handles page faults. It determines the address,
362 * and the problem, and then passes it off to one of the appropriate
363 * routines.
364 *
365 * interruption code (int_code):
366 * 04 Protection -> Write-Protection (suprression)
367 * 10 Segment translation -> Not present (nullification)
368 * 11 Page translation -> Not present (nullification)
369 * 3b Region third trans. -> Not present (nullification)
370 */
371static inline int do_exception(struct pt_regs *regs, int access)
372{
373#ifdef CONFIG_PGSTE
374 struct gmap *gmap;
375#endif
376 struct task_struct *tsk;
377 struct mm_struct *mm;
378 struct vm_area_struct *vma;
379 unsigned long trans_exc_code;
380 unsigned long address;
381 unsigned int flags;
382 int fault;
383
384 tsk = current;
385 /*
386 * The instruction that caused the program check has
387 * been nullified. Don't signal single step via SIGTRAP.
388 */
389 clear_pt_regs_flag(regs, PIF_PER_TRAP);
390
391 if (notify_page_fault(regs))
392 return 0;
393
394 mm = tsk->mm;
395 trans_exc_code = regs->int_parm_long;
396
397 /*
398 * Verify that the fault happened in user space, that
399 * we are not in an interrupt and that there is a
400 * user context.
401 */
402 fault = VM_FAULT_BADCONTEXT;
403 if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm))
404 goto out;
405
406 address = trans_exc_code & __FAIL_ADDR_MASK;
407 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
408 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
409 if (user_mode(regs))
410 flags |= FAULT_FLAG_USER;
411 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
412 flags |= FAULT_FLAG_WRITE;
413 down_read(&mm->mmap_sem);
414
415#ifdef CONFIG_PGSTE
416 gmap = (current->flags & PF_VCPU) ?
417 (struct gmap *) S390_lowcore.gmap : NULL;
418 if (gmap) {
419 current->thread.gmap_addr = address;
420 address = __gmap_translate(gmap, address);
421 if (address == -EFAULT) {
422 fault = VM_FAULT_BADMAP;
423 goto out_up;
424 }
425 if (gmap->pfault_enabled)
426 flags |= FAULT_FLAG_RETRY_NOWAIT;
427 }
428#endif
429
430retry:
431 fault = VM_FAULT_BADMAP;
432 vma = find_vma(mm, address);
433 if (!vma)
434 goto out_up;
435
436 if (unlikely(vma->vm_start > address)) {
437 if (!(vma->vm_flags & VM_GROWSDOWN))
438 goto out_up;
439 if (expand_stack(vma, address))
440 goto out_up;
441 }
442
443 /*
444 * Ok, we have a good vm_area for this memory access, so
445 * we can handle it..
446 */
447 fault = VM_FAULT_BADACCESS;
448 if (unlikely(!(vma->vm_flags & access)))
449 goto out_up;
450
451 if (is_vm_hugetlb_page(vma))
452 address &= HPAGE_MASK;
453 /*
454 * If for any reason at all we couldn't handle the fault,
455 * make sure we exit gracefully rather than endlessly redo
456 * the fault.
457 */
458 fault = handle_mm_fault(mm, vma, address, flags);
459 /* No reason to continue if interrupted by SIGKILL. */
460 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
461 fault = VM_FAULT_SIGNAL;
462 goto out;
463 }
464 if (unlikely(fault & VM_FAULT_ERROR))
465 goto out_up;
466
467 /*
468 * Major/minor page fault accounting is only done on the
469 * initial attempt. If we go through a retry, it is extremely
470 * likely that the page will be found in page cache at that point.
471 */
472 if (flags & FAULT_FLAG_ALLOW_RETRY) {
473 if (fault & VM_FAULT_MAJOR) {
474 tsk->maj_flt++;
475 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
476 regs, address);
477 } else {
478 tsk->min_flt++;
479 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
480 regs, address);
481 }
482 if (fault & VM_FAULT_RETRY) {
483#ifdef CONFIG_PGSTE
484 if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
485 /* FAULT_FLAG_RETRY_NOWAIT has been set,
486 * mmap_sem has not been released */
487 current->thread.gmap_pfault = 1;
488 fault = VM_FAULT_PFAULT;
489 goto out_up;
490 }
491#endif
492 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
493 * of starvation. */
494 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
495 FAULT_FLAG_RETRY_NOWAIT);
496 flags |= FAULT_FLAG_TRIED;
497 down_read(&mm->mmap_sem);
498 goto retry;
499 }
500 }
501#ifdef CONFIG_PGSTE
502 if (gmap) {
503 address = __gmap_link(gmap, current->thread.gmap_addr,
504 address);
505 if (address == -EFAULT) {
506 fault = VM_FAULT_BADMAP;
507 goto out_up;
508 }
509 if (address == -ENOMEM) {
510 fault = VM_FAULT_OOM;
511 goto out_up;
512 }
513 }
514#endif
515 fault = 0;
516out_up:
517 up_read(&mm->mmap_sem);
518out:
519 return fault;
520}
521
522void do_protection_exception(struct pt_regs *regs)
523{
524 unsigned long trans_exc_code;
525 int fault;
526
527 trans_exc_code = regs->int_parm_long;
528 /*
529 * Protection exceptions are suppressing, decrement psw address.
530 * The exception to this rule are aborted transactions, for these
531 * the PSW already points to the correct location.
532 */
533 if (!(regs->int_code & 0x200))
534 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
535 /*
536 * Check for low-address protection. This needs to be treated
537 * as a special case because the translation exception code
538 * field is not guaranteed to contain valid data in this case.
539 */
540 if (unlikely(!(trans_exc_code & 4))) {
541 do_low_address(regs);
542 return;
543 }
544 fault = do_exception(regs, VM_WRITE);
545 if (unlikely(fault))
546 do_fault_error(regs, fault);
547}
548NOKPROBE_SYMBOL(do_protection_exception);
549
550void do_dat_exception(struct pt_regs *regs)
551{
552 int access, fault;
553
554 access = VM_READ | VM_EXEC | VM_WRITE;
555 fault = do_exception(regs, access);
556 if (unlikely(fault))
557 do_fault_error(regs, fault);
558}
559NOKPROBE_SYMBOL(do_dat_exception);
560
561#ifdef CONFIG_PFAULT
562/*
563 * 'pfault' pseudo page faults routines.
564 */
565static int pfault_disable;
566
567static int __init nopfault(char *str)
568{
569 pfault_disable = 1;
570 return 1;
571}
572
573__setup("nopfault", nopfault);
574
575struct pfault_refbk {
576 u16 refdiagc;
577 u16 reffcode;
578 u16 refdwlen;
579 u16 refversn;
580 u64 refgaddr;
581 u64 refselmk;
582 u64 refcmpmk;
583 u64 reserved;
584} __attribute__ ((packed, aligned(8)));
585
586int pfault_init(void)
587{
588 struct pfault_refbk refbk = {
589 .refdiagc = 0x258,
590 .reffcode = 0,
591 .refdwlen = 5,
592 .refversn = 2,
593 .refgaddr = __LC_LPP,
594 .refselmk = 1ULL << 48,
595 .refcmpmk = 1ULL << 48,
596 .reserved = __PF_RES_FIELD };
597 int rc;
598
599 if (pfault_disable)
600 return -1;
601 diag_stat_inc(DIAG_STAT_X258);
602 asm volatile(
603 " diag %1,%0,0x258\n"
604 "0: j 2f\n"
605 "1: la %0,8\n"
606 "2:\n"
607 EX_TABLE(0b,1b)
608 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
609 return rc;
610}
611
612void pfault_fini(void)
613{
614 struct pfault_refbk refbk = {
615 .refdiagc = 0x258,
616 .reffcode = 1,
617 .refdwlen = 5,
618 .refversn = 2,
619 };
620
621 if (pfault_disable)
622 return;
623 diag_stat_inc(DIAG_STAT_X258);
624 asm volatile(
625 " diag %0,0,0x258\n"
626 "0:\n"
627 EX_TABLE(0b,0b)
628 : : "a" (&refbk), "m" (refbk) : "cc");
629}
630
631static DEFINE_SPINLOCK(pfault_lock);
632static LIST_HEAD(pfault_list);
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
643 * initial/completion signal bit. VM stores this
644 * in the 'cpu address' field associated with the
645 * external interrupt.
646 */
647 subcode = ext_code.subcode;
648 if ((subcode & 0xff00) != __SUBCODE_MASK)
649 return;
650 inc_irq_stat(IRQEXT_PFL);
651 /* Get the token (= pid of the affected task). */
652 pid = param64 & LPP_PFAULT_PID_MASK;
653 rcu_read_lock();
654 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
655 if (tsk)
656 get_task_struct(tsk);
657 rcu_read_unlock();
658 if (!tsk)
659 return;
660 spin_lock(&pfault_lock);
661 if (subcode & 0x0080) {
662 /* signal bit is set -> a page has been swapped in by VM */
663 if (tsk->thread.pfault_wait == 1) {
664 /* Initial interrupt was faster than the completion
665 * interrupt. pfault_wait is valid. Set pfault_wait
666 * back to zero and wake up the process. This can
667 * safely be done because the task is still sleeping
668 * and can't produce new pfaults. */
669 tsk->thread.pfault_wait = 0;
670 list_del(&tsk->thread.list);
671 wake_up_process(tsk);
672 put_task_struct(tsk);
673 } else {
674 /* Completion interrupt was faster than initial
675 * interrupt. Set pfault_wait to -1 so the initial
676 * interrupt doesn't put the task to sleep.
677 * If the task is not running, ignore the completion
678 * interrupt since it must be a leftover of a PFAULT
679 * CANCEL operation which didn't remove all pending
680 * completion interrupts. */
681 if (tsk->state == TASK_RUNNING)
682 tsk->thread.pfault_wait = -1;
683 }
684 } else {
685 /* signal bit not set -> a real page is missing. */
686 if (WARN_ON_ONCE(tsk != current))
687 goto out;
688 if (tsk->thread.pfault_wait == 1) {
689 /* Already on the list with a reference: put to sleep */
690 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
691 set_tsk_need_resched(tsk);
692 } else if (tsk->thread.pfault_wait == -1) {
693 /* Completion interrupt was faster than the initial
694 * interrupt (pfault_wait == -1). Set pfault_wait
695 * back to zero and exit. */
696 tsk->thread.pfault_wait = 0;
697 } else {
698 /* Initial interrupt arrived before completion
699 * interrupt. Let the task sleep.
700 * An extra task reference is needed since a different
701 * cpu may set the task state to TASK_RUNNING again
702 * before the scheduler is reached. */
703 get_task_struct(tsk);
704 tsk->thread.pfault_wait = 1;
705 list_add(&tsk->thread.list, &pfault_list);
706 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
707 set_tsk_need_resched(tsk);
708 }
709 }
710out:
711 spin_unlock(&pfault_lock);
712 put_task_struct(tsk);
713}
714
715static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
716 void *hcpu)
717{
718 struct thread_struct *thread, *next;
719 struct task_struct *tsk;
720
721 switch (action & ~CPU_TASKS_FROZEN) {
722 case CPU_DEAD:
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 break;
733 default:
734 break;
735 }
736 return NOTIFY_OK;
737}
738
739static int __init pfault_irq_init(void)
740{
741 int rc;
742
743 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
744 if (rc)
745 goto out_extint;
746 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
747 if (rc)
748 goto out_pfault;
749 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
750 hotcpu_notifier(pfault_cpu_notify, 0);
751 return 0;
752
753out_pfault:
754 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
755out_extint:
756 pfault_disable = 1;
757 return rc;
758}
759early_initcall(pfault_irq_init);
760
761#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 */