Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
fault.c
Go to the documentation of this file.
1 /*
2  * S390 version
3  * Copyright IBM Corp. 1999
4  * Author(s): Hartmut Penner ([email protected])
5  * Ulrich Weigand ([email protected])
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/pgtable.h>
34 #include <asm/irq.h>
35 #include <asm/mmu_context.h>
36 #include <asm/facility.h>
37 #include "../kernel/entry.h"
38 
39 #ifndef CONFIG_64BIT
40 #define __FAIL_ADDR_MASK 0x7ffff000
41 #define __SUBCODE_MASK 0x0200
42 #define __PF_RES_FIELD 0ULL
43 #else /* CONFIG_64BIT */
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47 #endif /* CONFIG_64BIT */
48 
49 #define VM_FAULT_BADCONTEXT 0x010000
50 #define VM_FAULT_BADMAP 0x020000
51 #define VM_FAULT_BADACCESS 0x040000
52 #define VM_FAULT_SIGNAL 0x080000
53 
54 static unsigned long store_indication;
55 
56 void fault_init(void)
57 {
58  if (test_facility(2) && test_facility(75))
59  store_indication = 0xc00;
60 }
61 
62 static inline int notify_page_fault(struct pt_regs *regs)
63 {
64  int ret = 0;
65 
66  /* kprobe_running() needs smp_processor_id() */
67  if (kprobes_built_in() && !user_mode(regs)) {
69  if (kprobe_running() && kprobe_fault_handler(regs, 14))
70  ret = 1;
72  }
73  return ret;
74 }
75 
76 
77 /*
78  * Unlock any spinlocks which will prevent us from getting the
79  * message out.
80  */
82 {
83  if (yes) {
84  oops_in_progress = 1;
85  } else {
86  int loglevel_save = console_loglevel;
88  oops_in_progress = 0;
89  /*
90  * OK, the message is on the console. Now we call printk()
91  * without oops_in_progress set so that printk will give klogd
92  * a poke. Hold onto your hats...
93  */
94  console_loglevel = 15;
95  printk(" ");
96  console_loglevel = loglevel_save;
97  }
98 }
99 
100 /*
101  * Returns the address space associated with the fault.
102  * Returns 0 for kernel space and 1 for user space.
103  */
104 static inline int user_space_fault(unsigned long trans_exc_code)
105 {
106  /*
107  * The lowest two bits of the translation exception
108  * identification indicate which paging table was used.
109  */
110  trans_exc_code &= 3;
111  if (trans_exc_code == 2)
112  /* Access via secondary space, set_fs setting decides */
113  return current->thread.mm_segment.ar4;
115  /* User space if the access has been done via home space. */
116  return trans_exc_code == 3;
117  /*
118  * If the user space is not the home space the kernel runs in home
119  * space. Access via secondary space has already been covered,
120  * access via primary space or access register is from user space
121  * and access via home space is from the kernel.
122  */
123  return trans_exc_code != 3;
124 }
125 
126 static inline void report_user_fault(struct pt_regs *regs, long signr)
127 {
128  if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
129  return;
130  if (!unhandled_signal(current, signr))
131  return;
132  if (!printk_ratelimit())
133  return;
134  printk(KERN_ALERT "User process fault: interruption code 0x%X ",
135  regs->int_code);
136  print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
137  printk(KERN_CONT "\n");
138  printk(KERN_ALERT "failing address: %lX\n",
140  show_regs(regs);
141 }
142 
143 /*
144  * Send SIGSEGV to task. This is an external routine
145  * to keep the stack usage of do_page_fault small.
146  */
147 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
148 {
149  struct siginfo si;
150 
151  report_user_fault(regs, SIGSEGV);
152  si.si_signo = SIGSEGV;
153  si.si_code = si_code;
154  si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
156 }
157 
158 static noinline void do_no_context(struct pt_regs *regs)
159 {
160  const struct exception_table_entry *fixup;
161  unsigned long address;
162 
163  /* Are we prepared to handle this kernel fault? */
164  fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
165  if (fixup) {
166  regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE;
167  return;
168  }
169 
170  /*
171  * Oops. The kernel tried to access some bad page. We'll have to
172  * terminate things with extreme prejudice.
173  */
174  address = regs->int_parm_long & __FAIL_ADDR_MASK;
175  if (!user_space_fault(regs->int_parm_long))
176  printk(KERN_ALERT "Unable to handle kernel pointer dereference"
177  " at virtual kernel address %p\n", (void *)address);
178  else
179  printk(KERN_ALERT "Unable to handle kernel paging request"
180  " at virtual user address %p\n", (void *)address);
181 
182  die(regs, "Oops");
183  do_exit(SIGKILL);
184 }
185 
186 static noinline void do_low_address(struct pt_regs *regs)
187 {
188  /* Low-address protection hit in kernel mode means
189  NULL pointer write access in kernel mode. */
190  if (regs->psw.mask & PSW_MASK_PSTATE) {
191  /* Low-address protection hit in user mode 'cannot happen'. */
192  die (regs, "Low-address protection");
193  do_exit(SIGKILL);
194  }
195 
196  do_no_context(regs);
197 }
198 
199 static noinline void do_sigbus(struct pt_regs *regs)
200 {
201  struct task_struct *tsk = current;
202  struct siginfo si;
203 
204  /*
205  * Send a sigbus, regardless of whether we were in kernel
206  * or user mode.
207  */
208  si.si_signo = SIGBUS;
209  si.si_errno = 0;
210  si.si_code = BUS_ADRERR;
211  si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
212  force_sig_info(SIGBUS, &si, tsk);
213 }
214 
215 static noinline void do_fault_error(struct pt_regs *regs, int fault)
216 {
217  int si_code;
218 
219  switch (fault) {
220  case VM_FAULT_BADACCESS:
221  case VM_FAULT_BADMAP:
222  /* Bad memory access. Check if it is kernel or user space. */
223  if (user_mode(regs)) {
224  /* User mode accesses just cause a SIGSEGV */
225  si_code = (fault == VM_FAULT_BADMAP) ?
227  do_sigsegv(regs, si_code);
228  return;
229  }
230  case VM_FAULT_BADCONTEXT:
231  do_no_context(regs);
232  break;
233  case VM_FAULT_SIGNAL:
234  if (!user_mode(regs))
235  do_no_context(regs);
236  break;
237  default: /* fault & VM_FAULT_ERROR */
238  if (fault & VM_FAULT_OOM) {
239  if (!user_mode(regs))
240  do_no_context(regs);
241  else
243  } else if (fault & VM_FAULT_SIGBUS) {
244  /* Kernel mode? Handle exceptions or die */
245  if (!user_mode(regs))
246  do_no_context(regs);
247  else
248  do_sigbus(regs);
249  } else
250  BUG();
251  break;
252  }
253 }
254 
255 /*
256  * This routine handles page faults. It determines the address,
257  * and the problem, and then passes it off to one of the appropriate
258  * routines.
259  *
260  * interruption code (int_code):
261  * 04 Protection -> Write-Protection (suprression)
262  * 10 Segment translation -> Not present (nullification)
263  * 11 Page translation -> Not present (nullification)
264  * 3b Region third trans. -> Not present (nullification)
265  */
266 static inline int do_exception(struct pt_regs *regs, int access)
267 {
268  struct task_struct *tsk;
269  struct mm_struct *mm;
270  struct vm_area_struct *vma;
271  unsigned long trans_exc_code;
272  unsigned long address;
273  unsigned int flags;
274  int fault;
275 
276  if (notify_page_fault(regs))
277  return 0;
278 
279  tsk = current;
280  mm = tsk->mm;
281  trans_exc_code = regs->int_parm_long;
282 
283  /*
284  * Verify that the fault happened in user space, that
285  * we are not in an interrupt and that there is a
286  * user context.
287  */
288  fault = VM_FAULT_BADCONTEXT;
289  if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
290  goto out;
291 
292  address = trans_exc_code & __FAIL_ADDR_MASK;
293  perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
294  flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
295  if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
296  flags |= FAULT_FLAG_WRITE;
297  down_read(&mm->mmap_sem);
298 
299 #ifdef CONFIG_PGSTE
300  if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
301  address = __gmap_fault(address,
302  (struct gmap *) S390_lowcore.gmap);
303  if (address == -EFAULT) {
304  fault = VM_FAULT_BADMAP;
305  goto out_up;
306  }
307  if (address == -ENOMEM) {
308  fault = VM_FAULT_OOM;
309  goto out_up;
310  }
311  }
312 #endif
313 
314 retry:
315  fault = VM_FAULT_BADMAP;
316  vma = find_vma(mm, address);
317  if (!vma)
318  goto out_up;
319 
320  if (unlikely(vma->vm_start > address)) {
321  if (!(vma->vm_flags & VM_GROWSDOWN))
322  goto out_up;
323  if (expand_stack(vma, address))
324  goto out_up;
325  }
326 
327  /*
328  * Ok, we have a good vm_area for this memory access, so
329  * we can handle it..
330  */
331  fault = VM_FAULT_BADACCESS;
332  if (unlikely(!(vma->vm_flags & access)))
333  goto out_up;
334 
335  if (is_vm_hugetlb_page(vma))
336  address &= HPAGE_MASK;
337  /*
338  * If for any reason at all we couldn't handle the fault,
339  * make sure we exit gracefully rather than endlessly redo
340  * the fault.
341  */
342  fault = handle_mm_fault(mm, vma, address, flags);
343  /* No reason to continue if interrupted by SIGKILL. */
344  if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
345  fault = VM_FAULT_SIGNAL;
346  goto out;
347  }
348  if (unlikely(fault & VM_FAULT_ERROR))
349  goto out_up;
350 
351  /*
352  * Major/minor page fault accounting is only done on the
353  * initial attempt. If we go through a retry, it is extremely
354  * likely that the page will be found in page cache at that point.
355  */
356  if (flags & FAULT_FLAG_ALLOW_RETRY) {
357  if (fault & VM_FAULT_MAJOR) {
358  tsk->maj_flt++;
359  perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
360  regs, address);
361  } else {
362  tsk->min_flt++;
363  perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
364  regs, address);
365  }
366  if (fault & VM_FAULT_RETRY) {
367  /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
368  * of starvation. */
369  flags &= ~FAULT_FLAG_ALLOW_RETRY;
370  flags |= FAULT_FLAG_TRIED;
371  down_read(&mm->mmap_sem);
372  goto retry;
373  }
374  }
375  /*
376  * The instruction that caused the program check will
377  * be repeated. Don't signal single step via SIGTRAP.
378  */
379  clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
380  fault = 0;
381 out_up:
382  up_read(&mm->mmap_sem);
383 out:
384  return fault;
385 }
386 
388 {
389  unsigned long trans_exc_code;
390  int fault;
391 
392  trans_exc_code = regs->int_parm_long;
393  /* Protection exception is suppressing, decrement psw address. */
394  regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
395  /*
396  * Check for low-address protection. This needs to be treated
397  * as a special case because the translation exception code
398  * field is not guaranteed to contain valid data in this case.
399  */
400  if (unlikely(!(trans_exc_code & 4))) {
401  do_low_address(regs);
402  return;
403  }
404  fault = do_exception(regs, VM_WRITE);
405  if (unlikely(fault))
406  do_fault_error(regs, fault);
407 }
408 
410 {
411  int access, fault;
412 
413  access = VM_READ | VM_EXEC | VM_WRITE;
414  fault = do_exception(regs, access);
415  if (unlikely(fault))
416  do_fault_error(regs, fault);
417 }
418 
419 #ifdef CONFIG_64BIT
420 void __kprobes do_asce_exception(struct pt_regs *regs)
421 {
422  struct mm_struct *mm = current->mm;
423  struct vm_area_struct *vma;
424  unsigned long trans_exc_code;
425 
426  trans_exc_code = regs->int_parm_long;
427  if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
428  goto no_context;
429 
430  down_read(&mm->mmap_sem);
431  vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
432  up_read(&mm->mmap_sem);
433 
434  if (vma) {
435  update_mm(mm, current);
436  return;
437  }
438 
439  /* User mode accesses just cause a SIGSEGV */
440  if (user_mode(regs)) {
441  do_sigsegv(regs, SEGV_MAPERR);
442  return;
443  }
444 
445 no_context:
446  do_no_context(regs);
447 }
448 #endif
449 
450 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
451 {
452  struct pt_regs regs;
453  int access, fault;
454 
455  /* Emulate a uaccess fault from kernel mode. */
457  if (!irqs_disabled())
458  regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
459  regs.psw.addr = (unsigned long) __builtin_return_address(0);
460  regs.psw.addr |= PSW_ADDR_AMODE;
461  regs.int_code = pgm_int_code;
462  regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
463  access = write ? VM_WRITE : VM_READ;
464  fault = do_exception(&regs, access);
465  /*
466  * Since the fault happened in kernel mode while performing a uaccess
467  * all we need to do now is emulating a fixup in case "fault" is not
468  * zero.
469  * For the calling uaccess functions this results always in -EFAULT.
470  */
471  return fault ? -EFAULT : 0;
472 }
473 
474 #ifdef CONFIG_PFAULT
475 /*
476  * 'pfault' pseudo page faults routines.
477  */
478 static int pfault_disable;
479 
480 static int __init nopfault(char *str)
481 {
482  pfault_disable = 1;
483  return 1;
484 }
485 
486 __setup("nopfault", nopfault);
487 
488 struct 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 
499 int 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 (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 
524 void 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 (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 
542 static DEFINE_SPINLOCK(pfault_lock);
543 static LIST_HEAD(pfault_list);
544 
545 static void pfault_interrupt(struct ext_code ext_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_code.subcode;
559  if ((subcode & 0xff00) != __SUBCODE_MASK)
560  return;
562  /* Get the token (= pid of the affected task). */
563  pid = sizeof(void *) == 4 ? param32 : param64;
564  rcu_read_lock();
565  tsk = find_task_by_pid_ns(pid, &init_pid_ns);
566  if (tsk)
567  get_task_struct(tsk);
568  rcu_read_unlock();
569  if (!tsk)
570  return;
571  spin_lock(&pfault_lock);
572  if (subcode & 0x0080) {
573  /* signal bit is set -> a page has been swapped in by VM */
574  if (tsk->thread.pfault_wait == 1) {
575  /* Initial interrupt was faster than the completion
576  * interrupt. pfault_wait is valid. Set pfault_wait
577  * back to zero and wake up the process. This can
578  * safely be done because the task is still sleeping
579  * and can't produce new pfaults. */
580  tsk->thread.pfault_wait = 0;
581  list_del(&tsk->thread.list);
582  wake_up_process(tsk);
583  put_task_struct(tsk);
584  } else {
585  /* Completion interrupt was faster than initial
586  * interrupt. Set pfault_wait to -1 so the initial
587  * interrupt doesn't put the task to sleep.
588  * If the task is not running, ignore the completion
589  * interrupt since it must be a leftover of a PFAULT
590  * CANCEL operation which didn't remove all pending
591  * completion interrupts. */
592  if (tsk->state == TASK_RUNNING)
593  tsk->thread.pfault_wait = -1;
594  }
595  } else {
596  /* signal bit not set -> a real page is missing. */
597  if (WARN_ON_ONCE(tsk != current))
598  goto out;
599  if (tsk->thread.pfault_wait == 1) {
600  /* Already on the list with a reference: put to sleep */
602  set_tsk_need_resched(tsk);
603  } else if (tsk->thread.pfault_wait == -1) {
604  /* Completion interrupt was faster than the initial
605  * interrupt (pfault_wait == -1). Set pfault_wait
606  * back to zero and exit. */
607  tsk->thread.pfault_wait = 0;
608  } else {
609  /* Initial interrupt arrived before completion
610  * interrupt. Let the task sleep.
611  * An extra task reference is needed since a different
612  * cpu may set the task state to TASK_RUNNING again
613  * before the scheduler is reached. */
614  get_task_struct(tsk);
615  tsk->thread.pfault_wait = 1;
616  list_add(&tsk->thread.list, &pfault_list);
618  set_tsk_need_resched(tsk);
619  }
620  }
621 out:
622  spin_unlock(&pfault_lock);
623  put_task_struct(tsk);
624 }
625 
626 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
627  unsigned long action, void *hcpu)
628 {
629  struct thread_struct *thread, *next;
630  struct task_struct *tsk;
631 
632  switch (action & ~CPU_TASKS_FROZEN) {
633  case CPU_DEAD:
634  spin_lock_irq(&pfault_lock);
635  list_for_each_entry_safe(thread, next, &pfault_list, list) {
636  thread->pfault_wait = 0;
637  list_del(&thread->list);
638  tsk = container_of(thread, struct task_struct, thread);
639  wake_up_process(tsk);
640  put_task_struct(tsk);
641  }
642  spin_unlock_irq(&pfault_lock);
643  break;
644  default:
645  break;
646  }
647  return NOTIFY_OK;
648 }
649 
650 static int __init pfault_irq_init(void)
651 {
652  int rc;
653 
654  rc = register_external_interrupt(0x2603, pfault_interrupt);
655  if (rc)
656  goto out_extint;
657  rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
658  if (rc)
659  goto out_pfault;
661  hotcpu_notifier(pfault_cpu_notify, 0);
662  return 0;
663 
664 out_pfault:
665  unregister_external_interrupt(0x2603, pfault_interrupt);
666 out_extint:
667  pfault_disable = 1;
668  return rc;
669 }
670 early_initcall(pfault_irq_init);
671 
672 #endif /* CONFIG_PFAULT */