1 /*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 1995 - 2000 by Ralf Baechle
7 */
8 #include <linux/context_tracking.h>
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/interrupt.h>
12 #include <linux/kernel.h>
13 #include <linux/errno.h>
14 #include <linux/string.h>
15 #include <linux/types.h>
16 #include <linux/ptrace.h>
17 #include <linux/ratelimit.h>
18 #include <linux/mman.h>
19 #include <linux/mm.h>
20 #include <linux/smp.h>
21 #include <linux/kprobes.h>
22 #include <linux/perf_event.h>
23 #include <linux/uaccess.h>
24
25 #include <asm/branch.h>
26 #include <asm/mmu_context.h>
27 #include <asm/ptrace.h>
28 #include <asm/highmem.h> /* For VMALLOC_END */
29 #include <linux/kdebug.h>
30
31 int show_unhandled_signals = 1;
32
33 /*
34 * This routine handles page faults. It determines the address,
35 * and the problem, and then passes it off to one of the appropriate
36 * routines.
37 */
__do_page_fault(struct pt_regs * regs,unsigned long write,unsigned long address)38 static void __do_page_fault(struct pt_regs *regs, unsigned long write,
39 unsigned long address)
40 {
41 struct vm_area_struct * vma = NULL;
42 struct task_struct *tsk = current;
43 struct mm_struct *mm = tsk->mm;
44 const int field = sizeof(unsigned long) * 2;
45 int si_code;
46 vm_fault_t fault;
47 unsigned int flags = FAULT_FLAG_DEFAULT;
48
49 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
50
51 #if 0
52 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
53 current->comm, current->pid, field, address, write,
54 field, regs->cp0_epc);
55 #endif
56
57 #ifdef CONFIG_KPROBES
58 /*
59 * This is to notify the fault handler of the kprobes.
60 */
61 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
62 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
63 return;
64 #endif
65
66 si_code = SEGV_MAPERR;
67
68 /*
69 * We fault-in kernel-space virtual memory on-demand. The
70 * 'reference' page table is init_mm.pgd.
71 *
72 * NOTE! We MUST NOT take any locks for this case. We may
73 * be in an interrupt or a critical region, and should
74 * only copy the information from the master page table,
75 * nothing more.
76 */
77 #ifdef CONFIG_64BIT
78 # define VMALLOC_FAULT_TARGET no_context
79 #else
80 # define VMALLOC_FAULT_TARGET vmalloc_fault
81 #endif
82
83 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
84 goto VMALLOC_FAULT_TARGET;
85 #ifdef MODULE_START
86 if (unlikely(address >= MODULE_START && address < MODULE_END))
87 goto VMALLOC_FAULT_TARGET;
88 #endif
89
90 /*
91 * If we're in an interrupt or have no user
92 * context, we must not take the fault..
93 */
94 if (faulthandler_disabled() || !mm)
95 goto bad_area_nosemaphore;
96
97 if (user_mode(regs))
98 flags |= FAULT_FLAG_USER;
99
100 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
101 retry:
102 vma = lock_mm_and_find_vma(mm, address, regs);
103 if (!vma)
104 goto bad_area_nosemaphore;
105 /*
106 * Ok, we have a good vm_area for this memory access, so
107 * we can handle it..
108 */
109 si_code = SEGV_ACCERR;
110
111 if (write) {
112 if (!(vma->vm_flags & VM_WRITE))
113 goto bad_area;
114 flags |= FAULT_FLAG_WRITE;
115 } else {
116 if (cpu_has_rixi) {
117 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
118 #if 0
119 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
120 raw_smp_processor_id(),
121 current->comm, current->pid,
122 field, address, write,
123 field, regs->cp0_epc);
124 #endif
125 goto bad_area;
126 }
127 if (!(vma->vm_flags & VM_READ) &&
128 exception_epc(regs) != address) {
129 #if 0
130 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
131 raw_smp_processor_id(),
132 current->comm, current->pid,
133 field, address, write,
134 field, regs->cp0_epc);
135 #endif
136 goto bad_area;
137 }
138 } else {
139 if (unlikely(!vma_is_accessible(vma)))
140 goto bad_area;
141 }
142 }
143
144 /*
145 * If for any reason at all we couldn't handle the fault,
146 * make sure we exit gracefully rather than endlessly redo
147 * the fault.
148 */
149 fault = handle_mm_fault(vma, address, flags, regs);
150
151 if (fault_signal_pending(fault, regs)) {
152 if (!user_mode(regs))
153 goto no_context;
154 return;
155 }
156
157 /* The fault is fully completed (including releasing mmap lock) */
158 if (fault & VM_FAULT_COMPLETED)
159 return;
160
161 if (unlikely(fault & VM_FAULT_ERROR)) {
162 if (fault & VM_FAULT_OOM)
163 goto out_of_memory;
164 else if (fault & VM_FAULT_SIGSEGV)
165 goto bad_area;
166 else if (fault & VM_FAULT_SIGBUS)
167 goto do_sigbus;
168 BUG();
169 }
170
171 if (fault & VM_FAULT_RETRY) {
172 flags |= FAULT_FLAG_TRIED;
173
174 /*
175 * No need to mmap_read_unlock(mm) as we would
176 * have already released it in __lock_page_or_retry
177 * in mm/filemap.c.
178 */
179
180 goto retry;
181 }
182
183 mmap_read_unlock(mm);
184 return;
185
186 /*
187 * Something tried to access memory that isn't in our memory map..
188 * Fix it, but check if it's kernel or user first..
189 */
190 bad_area:
191 mmap_read_unlock(mm);
192
193 bad_area_nosemaphore:
194 /* User mode accesses just cause a SIGSEGV */
195 if (user_mode(regs)) {
196 tsk->thread.cp0_badvaddr = address;
197 tsk->thread.error_code = write;
198 if (show_unhandled_signals &&
199 unhandled_signal(tsk, SIGSEGV) &&
200 __ratelimit(&ratelimit_state)) {
201 pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n",
202 tsk->comm,
203 write ? "write access to" : "read access from",
204 field, address);
205 pr_info("epc = %0*lx in", field,
206 (unsigned long) regs->cp0_epc);
207 print_vma_addr(KERN_CONT " ", regs->cp0_epc);
208 pr_cont("\n");
209 pr_info("ra = %0*lx in", field,
210 (unsigned long) regs->regs[31]);
211 print_vma_addr(KERN_CONT " ", regs->regs[31]);
212 pr_cont("\n");
213 }
214 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
215 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
216 return;
217 }
218
219 no_context:
220 /* Are we prepared to handle this kernel fault? */
221 if (fixup_exception(regs)) {
222 current->thread.cp0_baduaddr = address;
223 return;
224 }
225
226 /*
227 * Oops. The kernel tried to access some bad page. We'll have to
228 * terminate things with extreme prejudice.
229 */
230 bust_spinlocks(1);
231
232 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
233 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
234 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
235 field, regs->regs[31]);
236 die("Oops", regs);
237
238 out_of_memory:
239 /*
240 * We ran out of memory, call the OOM killer, and return the userspace
241 * (which will retry the fault, or kill us if we got oom-killed).
242 */
243 mmap_read_unlock(mm);
244 if (!user_mode(regs))
245 goto no_context;
246 pagefault_out_of_memory();
247 return;
248
249 do_sigbus:
250 mmap_read_unlock(mm);
251
252 /* Kernel mode? Handle exceptions or die */
253 if (!user_mode(regs))
254 goto no_context;
255
256 /*
257 * Send a sigbus, regardless of whether we were in kernel
258 * or user mode.
259 */
260 #if 0
261 printk("do_page_fault() #3: sending SIGBUS to %s for "
262 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
263 tsk->comm,
264 write ? "write access to" : "read access from",
265 field, address,
266 field, (unsigned long) regs->cp0_epc,
267 field, (unsigned long) regs->regs[31]);
268 #endif
269 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
270 tsk->thread.cp0_badvaddr = address;
271 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
272
273 return;
274 #ifndef CONFIG_64BIT
275 vmalloc_fault:
276 {
277 /*
278 * Synchronize this task's top level page-table
279 * with the 'reference' page table.
280 *
281 * Do _not_ use "tsk" here. We might be inside
282 * an interrupt in the middle of a task switch..
283 */
284 int offset = pgd_index(address);
285 pgd_t *pgd, *pgd_k;
286 p4d_t *p4d, *p4d_k;
287 pud_t *pud, *pud_k;
288 pmd_t *pmd, *pmd_k;
289 pte_t *pte_k;
290
291 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
292 pgd_k = init_mm.pgd + offset;
293
294 if (!pgd_present(*pgd_k))
295 goto no_context;
296 set_pgd(pgd, *pgd_k);
297
298 p4d = p4d_offset(pgd, address);
299 p4d_k = p4d_offset(pgd_k, address);
300 if (!p4d_present(*p4d_k))
301 goto no_context;
302
303 pud = pud_offset(p4d, address);
304 pud_k = pud_offset(p4d_k, address);
305 if (!pud_present(*pud_k))
306 goto no_context;
307
308 pmd = pmd_offset(pud, address);
309 pmd_k = pmd_offset(pud_k, address);
310 if (!pmd_present(*pmd_k))
311 goto no_context;
312 set_pmd(pmd, *pmd_k);
313
314 pte_k = pte_offset_kernel(pmd_k, address);
315 if (!pte_present(*pte_k))
316 goto no_context;
317 return;
318 }
319 #endif
320 }
321 NOKPROBE_SYMBOL(__do_page_fault);
322
do_page_fault(struct pt_regs * regs,unsigned long write,unsigned long address)323 asmlinkage void do_page_fault(struct pt_regs *regs,
324 unsigned long write, unsigned long address)
325 {
326 enum ctx_state prev_state;
327
328 prev_state = exception_enter();
329 __do_page_fault(regs, write, address);
330 exception_exit(prev_state);
331 }
332 NOKPROBE_SYMBOL(do_page_fault);
333