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  * OpenRISC fault.c
3  *
4  * Linux architectural port borrowing liberally from similar works of
5  * others. All original copyrights apply as per the original source
6  * declaration.
7  *
8  * Modifications for the OpenRISC architecture:
9  * Copyright (C) 2003 Matjaz Breskvar <[email protected]>
10  * Copyright (C) 2010-2011 Jonas Bonn <[email protected]>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public License
14  * as published by the Free Software Foundation; either version
15  * 2 of the License, or (at your option) any later version.
16  */
17 
18 #include <linux/mm.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/sched.h>
22 
23 #include <asm/uaccess.h>
24 #include <asm/siginfo.h>
25 #include <asm/signal.h>
26 
27 #define NUM_TLB_ENTRIES 64
28 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
29 
30 unsigned long pte_misses; /* updated by do_page_fault() */
31 unsigned long pte_errors; /* updated by do_page_fault() */
32 
33 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
34  * - also look into include/asm-or32/mmu_context.h
35  */
36 volatile pgd_t *current_pgd;
37 
38 extern void die(char *, struct pt_regs *, long);
39 
40 /*
41  * This routine handles page faults. It determines the address,
42  * and the problem, and then passes it off to one of the appropriate
43  * routines.
44  *
45  * If this routine detects a bad access, it returns 1, otherwise it
46  * returns 0.
47  */
48 
49 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
50  unsigned long vector, int write_acc)
51 {
52  struct task_struct *tsk;
53  struct mm_struct *mm;
54  struct vm_area_struct *vma;
56  int fault;
57  unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
58 
59  tsk = current;
60 
61  /*
62  * We fault-in kernel-space virtual memory on-demand. The
63  * 'reference' page table is init_mm.pgd.
64  *
65  * NOTE! We MUST NOT take any locks for this case. We may
66  * be in an interrupt or a critical region, and should
67  * only copy the information from the master page table,
68  * nothing more.
69  *
70  * NOTE2: This is done so that, when updating the vmalloc
71  * mappings we don't have to walk all processes pgdirs and
72  * add the high mappings all at once. Instead we do it as they
73  * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
74  * bit set so sometimes the TLB can use a lingering entry.
75  *
76  * This verifies that the fault happens in kernel space
77  * and that the fault was not a protection error.
78  */
79 
80  if (address >= VMALLOC_START &&
81  (vector != 0x300 && vector != 0x400) &&
82  !user_mode(regs))
83  goto vmalloc_fault;
84 
85  /* If exceptions were enabled, we can reenable them here */
86  if (user_mode(regs)) {
87  /* Exception was in userspace: reenable interrupts */
89  } else {
90  /* If exception was in a syscall, then IRQ's may have
91  * been enabled or disabled. If they were enabled,
92  * reenable them.
93  */
94  if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
96  }
97 
98  mm = tsk->mm;
99  info.si_code = SEGV_MAPERR;
100 
101  /*
102  * If we're in an interrupt or have no user
103  * context, we must not take the fault..
104  */
105 
106  if (in_interrupt() || !mm)
107  goto no_context;
108 
109 retry:
110  down_read(&mm->mmap_sem);
111  vma = find_vma(mm, address);
112 
113  if (!vma)
114  goto bad_area;
115 
116  if (vma->vm_start <= address)
117  goto good_area;
118 
119  if (!(vma->vm_flags & VM_GROWSDOWN))
120  goto bad_area;
121 
122  if (user_mode(regs)) {
123  /*
124  * accessing the stack below usp is always a bug.
125  * we get page-aligned addresses so we can only check
126  * if we're within a page from usp, but that might be
127  * enough to catch brutal errors at least.
128  */
129  if (address + PAGE_SIZE < regs->sp)
130  goto bad_area;
131  }
132  if (expand_stack(vma, address))
133  goto bad_area;
134 
135  /*
136  * Ok, we have a good vm_area for this memory access, so
137  * we can handle it..
138  */
139 
140 good_area:
141  info.si_code = SEGV_ACCERR;
142 
143  /* first do some preliminary protection checks */
144 
145  if (write_acc) {
146  if (!(vma->vm_flags & VM_WRITE))
147  goto bad_area;
148  flags |= FAULT_FLAG_WRITE;
149  } else {
150  /* not present */
151  if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
152  goto bad_area;
153  }
154 
155  /* are we trying to execute nonexecutable area */
156  if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
157  goto bad_area;
158 
159  /*
160  * If for any reason at all we couldn't handle the fault,
161  * make sure we exit gracefully rather than endlessly redo
162  * the fault.
163  */
164 
165  fault = handle_mm_fault(mm, vma, address, flags);
166 
167  if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
168  return;
169 
170  if (unlikely(fault & VM_FAULT_ERROR)) {
171  if (fault & VM_FAULT_OOM)
172  goto out_of_memory;
173  else if (fault & VM_FAULT_SIGBUS)
174  goto do_sigbus;
175  BUG();
176  }
177 
178  if (flags & FAULT_FLAG_ALLOW_RETRY) {
179  /*RGD modeled on Cris */
180  if (fault & VM_FAULT_MAJOR)
181  tsk->maj_flt++;
182  else
183  tsk->min_flt++;
184  if (fault & VM_FAULT_RETRY) {
185  flags &= ~FAULT_FLAG_ALLOW_RETRY;
186  flags |= FAULT_FLAG_TRIED;
187 
188  /* No need to up_read(&mm->mmap_sem) as we would
189  * have already released it in __lock_page_or_retry
190  * in mm/filemap.c.
191  */
192 
193  goto retry;
194  }
195  }
196 
197  up_read(&mm->mmap_sem);
198  return;
199 
200  /*
201  * Something tried to access memory that isn't in our memory map..
202  * Fix it, but check if it's kernel or user first..
203  */
204 
205 bad_area:
206  up_read(&mm->mmap_sem);
207 
208 bad_area_nosemaphore:
209 
210  /* User mode accesses just cause a SIGSEGV */
211 
212  if (user_mode(regs)) {
213  info.si_signo = SIGSEGV;
214  info.si_errno = 0;
215  /* info.si_code has been set above */
216  info.si_addr = (void *)address;
217  force_sig_info(SIGSEGV, &info, tsk);
218  return;
219  }
220 
221 no_context:
222 
223  /* Are we prepared to handle this kernel fault?
224  *
225  * (The kernel has valid exception-points in the source
226  * when it acesses user-memory. When it fails in one
227  * of those points, we find it in a table and do a jump
228  * to some fixup code that loads an appropriate error
229  * code)
230  */
231 
232  {
233  const struct exception_table_entry *entry;
234 
235  __asm__ __volatile__("l.nop 42");
236 
237  if ((entry = search_exception_tables(regs->pc)) != NULL) {
238  /* Adjust the instruction pointer in the stackframe */
239  regs->pc = entry->fixup;
240  return;
241  }
242  }
243 
244  /*
245  * Oops. The kernel tried to access some bad page. We'll have to
246  * terminate things with extreme prejudice.
247  */
248 
249  if ((unsigned long)(address) < PAGE_SIZE)
251  "Unable to handle kernel NULL pointer dereference");
252  else
253  printk(KERN_ALERT "Unable to handle kernel access");
254  printk(" at virtual address 0x%08lx\n", address);
255 
256  die("Oops", regs, write_acc);
257 
258  do_exit(SIGKILL);
259 
260  /*
261  * We ran out of memory, or some other thing happened to us that made
262  * us unable to handle the page fault gracefully.
263  */
264 
266  __asm__ __volatile__("l.nop 42");
267  __asm__ __volatile__("l.nop 1");
268 
269  up_read(&mm->mmap_sem);
270  printk("VM: killing process %s\n", tsk->comm);
271  if (user_mode(regs))
272  do_exit(SIGKILL);
273  goto no_context;
274 
275 do_sigbus:
276  up_read(&mm->mmap_sem);
277 
278  /*
279  * Send a sigbus, regardless of whether we were in kernel
280  * or user mode.
281  */
282  info.si_signo = SIGBUS;
283  info.si_errno = 0;
284  info.si_code = BUS_ADRERR;
285  info.si_addr = (void *)address;
286  force_sig_info(SIGBUS, &info, tsk);
287 
288  /* Kernel mode? Handle exceptions or die */
289  if (!user_mode(regs))
290  goto no_context;
291  return;
292 
293 vmalloc_fault:
294  {
295  /*
296  * Synchronize this task's top level page-table
297  * with the 'reference' page table.
298  *
299  * Use current_pgd instead of tsk->active_mm->pgd
300  * since the latter might be unavailable if this
301  * code is executed in a misfortunately run irq
302  * (like inside schedule() between switch_mm and
303  * switch_to...).
304  */
305 
306  int offset = pgd_index(address);
307  pgd_t *pgd, *pgd_k;
308  pud_t *pud, *pud_k;
309  pmd_t *pmd, *pmd_k;
310  pte_t *pte_k;
311 
312 /*
313  phx_warn("do_page_fault(): vmalloc_fault will not work, "
314  "since current_pgd assign a proper value somewhere\n"
315  "anyhow we don't need this at the moment\n");
316 
317  phx_mmu("vmalloc_fault");
318 */
319  pgd = (pgd_t *)current_pgd + offset;
320  pgd_k = init_mm.pgd + offset;
321 
322  /* Since we're two-level, we don't need to do both
323  * set_pgd and set_pmd (they do the same thing). If
324  * we go three-level at some point, do the right thing
325  * with pgd_present and set_pgd here.
326  *
327  * Also, since the vmalloc area is global, we don't
328  * need to copy individual PTE's, it is enough to
329  * copy the pgd pointer into the pte page of the
330  * root task. If that is there, we'll find our pte if
331  * it exists.
332  */
333 
334  pud = pud_offset(pgd, address);
335  pud_k = pud_offset(pgd_k, address);
336  if (!pud_present(*pud_k))
337  goto no_context;
338 
339  pmd = pmd_offset(pud, address);
340  pmd_k = pmd_offset(pud_k, address);
341 
342  if (!pmd_present(*pmd_k))
343  goto bad_area_nosemaphore;
344 
345  set_pmd(pmd, *pmd_k);
346 
347  /* Make sure the actual PTE exists as well to
348  * catch kernel vmalloc-area accesses to non-mapped
349  * addresses. If we don't do this, this will just
350  * silently loop forever.
351  */
352 
353  pte_k = pte_offset_kernel(pmd_k, address);
354  if (!pte_present(*pte_k))
355  goto no_context;
356 
357  return;
358  }
359 }