Loading...
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
31int 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 */
38static 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);
101retry:
102 mmap_read_lock(mm);
103 vma = find_vma(mm, address);
104 if (!vma)
105 goto bad_area;
106 if (vma->vm_start <= address)
107 goto good_area;
108 if (!(vma->vm_flags & VM_GROWSDOWN))
109 goto bad_area;
110 if (expand_stack(vma, address))
111 goto bad_area;
112/*
113 * Ok, we have a good vm_area for this memory access, so
114 * we can handle it..
115 */
116good_area:
117 si_code = SEGV_ACCERR;
118
119 if (write) {
120 if (!(vma->vm_flags & VM_WRITE))
121 goto bad_area;
122 flags |= FAULT_FLAG_WRITE;
123 } else {
124 if (cpu_has_rixi) {
125 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
126#if 0
127 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
128 raw_smp_processor_id(),
129 current->comm, current->pid,
130 field, address, write,
131 field, regs->cp0_epc);
132#endif
133 goto bad_area;
134 }
135 if (!(vma->vm_flags & VM_READ) &&
136 exception_epc(regs) != address) {
137#if 0
138 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
139 raw_smp_processor_id(),
140 current->comm, current->pid,
141 field, address, write,
142 field, regs->cp0_epc);
143#endif
144 goto bad_area;
145 }
146 } else {
147 if (unlikely(!vma_is_accessible(vma)))
148 goto bad_area;
149 }
150 }
151
152 /*
153 * If for any reason at all we couldn't handle the fault,
154 * make sure we exit gracefully rather than endlessly redo
155 * the fault.
156 */
157 fault = handle_mm_fault(vma, address, flags, regs);
158
159 if (fault_signal_pending(fault, regs)) {
160 if (!user_mode(regs))
161 goto no_context;
162 return;
163 }
164
165 /* The fault is fully completed (including releasing mmap lock) */
166 if (fault & VM_FAULT_COMPLETED)
167 return;
168
169 if (unlikely(fault & VM_FAULT_ERROR)) {
170 if (fault & VM_FAULT_OOM)
171 goto out_of_memory;
172 else if (fault & VM_FAULT_SIGSEGV)
173 goto bad_area;
174 else if (fault & VM_FAULT_SIGBUS)
175 goto do_sigbus;
176 BUG();
177 }
178
179 if (fault & VM_FAULT_RETRY) {
180 flags |= FAULT_FLAG_TRIED;
181
182 /*
183 * No need to mmap_read_unlock(mm) as we would
184 * have already released it in __lock_page_or_retry
185 * in mm/filemap.c.
186 */
187
188 goto retry;
189 }
190
191 mmap_read_unlock(mm);
192 return;
193
194/*
195 * Something tried to access memory that isn't in our memory map..
196 * Fix it, but check if it's kernel or user first..
197 */
198bad_area:
199 mmap_read_unlock(mm);
200
201bad_area_nosemaphore:
202 /* User mode accesses just cause a SIGSEGV */
203 if (user_mode(regs)) {
204 tsk->thread.cp0_badvaddr = address;
205 tsk->thread.error_code = write;
206 if (show_unhandled_signals &&
207 unhandled_signal(tsk, SIGSEGV) &&
208 __ratelimit(&ratelimit_state)) {
209 pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n",
210 tsk->comm,
211 write ? "write access to" : "read access from",
212 field, address);
213 pr_info("epc = %0*lx in", field,
214 (unsigned long) regs->cp0_epc);
215 print_vma_addr(KERN_CONT " ", regs->cp0_epc);
216 pr_cont("\n");
217 pr_info("ra = %0*lx in", field,
218 (unsigned long) regs->regs[31]);
219 print_vma_addr(KERN_CONT " ", regs->regs[31]);
220 pr_cont("\n");
221 }
222 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
223 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
224 return;
225 }
226
227no_context:
228 /* Are we prepared to handle this kernel fault? */
229 if (fixup_exception(regs)) {
230 current->thread.cp0_baduaddr = address;
231 return;
232 }
233
234 /*
235 * Oops. The kernel tried to access some bad page. We'll have to
236 * terminate things with extreme prejudice.
237 */
238 bust_spinlocks(1);
239
240 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
241 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
242 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
243 field, regs->regs[31]);
244 die("Oops", regs);
245
246out_of_memory:
247 /*
248 * We ran out of memory, call the OOM killer, and return the userspace
249 * (which will retry the fault, or kill us if we got oom-killed).
250 */
251 mmap_read_unlock(mm);
252 if (!user_mode(regs))
253 goto no_context;
254 pagefault_out_of_memory();
255 return;
256
257do_sigbus:
258 mmap_read_unlock(mm);
259
260 /* Kernel mode? Handle exceptions or die */
261 if (!user_mode(regs))
262 goto no_context;
263
264 /*
265 * Send a sigbus, regardless of whether we were in kernel
266 * or user mode.
267 */
268#if 0
269 printk("do_page_fault() #3: sending SIGBUS to %s for "
270 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
271 tsk->comm,
272 write ? "write access to" : "read access from",
273 field, address,
274 field, (unsigned long) regs->cp0_epc,
275 field, (unsigned long) regs->regs[31]);
276#endif
277 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
278 tsk->thread.cp0_badvaddr = address;
279 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
280
281 return;
282#ifndef CONFIG_64BIT
283vmalloc_fault:
284 {
285 /*
286 * Synchronize this task's top level page-table
287 * with the 'reference' page table.
288 *
289 * Do _not_ use "tsk" here. We might be inside
290 * an interrupt in the middle of a task switch..
291 */
292 int offset = pgd_index(address);
293 pgd_t *pgd, *pgd_k;
294 p4d_t *p4d, *p4d_k;
295 pud_t *pud, *pud_k;
296 pmd_t *pmd, *pmd_k;
297 pte_t *pte_k;
298
299 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
300 pgd_k = init_mm.pgd + offset;
301
302 if (!pgd_present(*pgd_k))
303 goto no_context;
304 set_pgd(pgd, *pgd_k);
305
306 p4d = p4d_offset(pgd, address);
307 p4d_k = p4d_offset(pgd_k, address);
308 if (!p4d_present(*p4d_k))
309 goto no_context;
310
311 pud = pud_offset(p4d, address);
312 pud_k = pud_offset(p4d_k, address);
313 if (!pud_present(*pud_k))
314 goto no_context;
315
316 pmd = pmd_offset(pud, address);
317 pmd_k = pmd_offset(pud_k, address);
318 if (!pmd_present(*pmd_k))
319 goto no_context;
320 set_pmd(pmd, *pmd_k);
321
322 pte_k = pte_offset_kernel(pmd_k, address);
323 if (!pte_present(*pte_k))
324 goto no_context;
325 return;
326 }
327#endif
328}
329NOKPROBE_SYMBOL(__do_page_fault);
330
331asmlinkage void do_page_fault(struct pt_regs *regs,
332 unsigned long write, unsigned long address)
333{
334 enum ctx_state prev_state;
335
336 prev_state = exception_enter();
337 __do_page_fault(regs, write, address);
338 exception_exit(prev_state);
339}
340NOKPROBE_SYMBOL(do_page_fault);
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/signal.h>
9#include <linux/sched.h>
10#include <linux/interrupt.h>
11#include <linux/kernel.h>
12#include <linux/errno.h>
13#include <linux/string.h>
14#include <linux/types.h>
15#include <linux/ptrace.h>
16#include <linux/mman.h>
17#include <linux/mm.h>
18#include <linux/smp.h>
19#include <linux/module.h>
20#include <linux/kprobes.h>
21#include <linux/perf_event.h>
22
23#include <asm/branch.h>
24#include <asm/mmu_context.h>
25#include <asm/uaccess.h>
26#include <asm/ptrace.h>
27#include <asm/highmem.h> /* For VMALLOC_END */
28#include <linux/kdebug.h>
29
30/*
31 * This routine handles page faults. It determines the address,
32 * and the problem, and then passes it off to one of the appropriate
33 * routines.
34 */
35asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, unsigned long write,
36 unsigned long address)
37{
38 struct vm_area_struct * vma = NULL;
39 struct task_struct *tsk = current;
40 struct mm_struct *mm = tsk->mm;
41 const int field = sizeof(unsigned long) * 2;
42 siginfo_t info;
43 int fault;
44 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
45 (write ? FAULT_FLAG_WRITE : 0);
46
47#if 0
48 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
49 current->comm, current->pid, field, address, write,
50 field, regs->cp0_epc);
51#endif
52
53#ifdef CONFIG_KPROBES
54 /*
55 * This is to notify the fault handler of the kprobes. The
56 * exception code is redundant as it is also carried in REGS,
57 * but we pass it anyhow.
58 */
59 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
60 (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
61 return;
62#endif
63
64 info.si_code = SEGV_MAPERR;
65
66 /*
67 * We fault-in kernel-space virtual memory on-demand. The
68 * 'reference' page table is init_mm.pgd.
69 *
70 * NOTE! We MUST NOT take any locks for this case. We may
71 * be in an interrupt or a critical region, and should
72 * only copy the information from the master page table,
73 * nothing more.
74 */
75#ifdef CONFIG_64BIT
76# define VMALLOC_FAULT_TARGET no_context
77#else
78# define VMALLOC_FAULT_TARGET vmalloc_fault
79#endif
80
81 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
82 goto VMALLOC_FAULT_TARGET;
83#ifdef MODULE_START
84 if (unlikely(address >= MODULE_START && address < MODULE_END))
85 goto VMALLOC_FAULT_TARGET;
86#endif
87
88 /*
89 * If we're in an interrupt or have no user
90 * context, we must not take the fault..
91 */
92 if (in_atomic() || !mm)
93 goto bad_area_nosemaphore;
94
95retry:
96 down_read(&mm->mmap_sem);
97 vma = find_vma(mm, address);
98 if (!vma)
99 goto bad_area;
100 if (vma->vm_start <= address)
101 goto good_area;
102 if (!(vma->vm_flags & VM_GROWSDOWN))
103 goto bad_area;
104 if (expand_stack(vma, address))
105 goto bad_area;
106/*
107 * Ok, we have a good vm_area for this memory access, so
108 * we can handle it..
109 */
110good_area:
111 info.si_code = SEGV_ACCERR;
112
113 if (write) {
114 if (!(vma->vm_flags & VM_WRITE))
115 goto bad_area;
116 } else {
117 if (kernel_uses_smartmips_rixi) {
118 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
119#if 0
120 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
121 raw_smp_processor_id(),
122 current->comm, current->pid,
123 field, address, write,
124 field, regs->cp0_epc);
125#endif
126 goto bad_area;
127 }
128 if (!(vma->vm_flags & VM_READ)) {
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 (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
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(mm, vma, address, flags);
150
151 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
152 return;
153
154 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
155 if (unlikely(fault & VM_FAULT_ERROR)) {
156 if (fault & VM_FAULT_OOM)
157 goto out_of_memory;
158 else if (fault & VM_FAULT_SIGBUS)
159 goto do_sigbus;
160 BUG();
161 }
162 if (flags & FAULT_FLAG_ALLOW_RETRY) {
163 if (fault & VM_FAULT_MAJOR) {
164 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
165 regs, address);
166 tsk->maj_flt++;
167 } else {
168 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
169 regs, address);
170 tsk->min_flt++;
171 }
172 if (fault & VM_FAULT_RETRY) {
173 flags &= ~FAULT_FLAG_ALLOW_RETRY;
174
175 /*
176 * No need to up_read(&mm->mmap_sem) as we would
177 * have already released it in __lock_page_or_retry
178 * in mm/filemap.c.
179 */
180
181 goto retry;
182 }
183 }
184
185 up_read(&mm->mmap_sem);
186 return;
187
188/*
189 * Something tried to access memory that isn't in our memory map..
190 * Fix it, but check if it's kernel or user first..
191 */
192bad_area:
193 up_read(&mm->mmap_sem);
194
195bad_area_nosemaphore:
196 /* User mode accesses just cause a SIGSEGV */
197 if (user_mode(regs)) {
198 tsk->thread.cp0_badvaddr = address;
199 tsk->thread.error_code = write;
200#if 0
201 printk("do_page_fault() #2: sending SIGSEGV to %s for "
202 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
203 tsk->comm,
204 write ? "write access to" : "read access from",
205 field, address,
206 field, (unsigned long) regs->cp0_epc,
207 field, (unsigned long) regs->regs[31]);
208#endif
209 info.si_signo = SIGSEGV;
210 info.si_errno = 0;
211 /* info.si_code has been set above */
212 info.si_addr = (void __user *) address;
213 force_sig_info(SIGSEGV, &info, tsk);
214 return;
215 }
216
217no_context:
218 /* Are we prepared to handle this kernel fault? */
219 if (fixup_exception(regs)) {
220 current->thread.cp0_baduaddr = address;
221 return;
222 }
223
224 /*
225 * Oops. The kernel tried to access some bad page. We'll have to
226 * terminate things with extreme prejudice.
227 */
228 bust_spinlocks(1);
229
230 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
231 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
232 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
233 field, regs->regs[31]);
234 die("Oops", regs);
235
236out_of_memory:
237 /*
238 * We ran out of memory, call the OOM killer, and return the userspace
239 * (which will retry the fault, or kill us if we got oom-killed).
240 */
241 up_read(&mm->mmap_sem);
242 pagefault_out_of_memory();
243 return;
244
245do_sigbus:
246 up_read(&mm->mmap_sem);
247
248 /* Kernel mode? Handle exceptions or die */
249 if (!user_mode(regs))
250 goto no_context;
251 else
252 /*
253 * Send a sigbus, regardless of whether we were in kernel
254 * or user mode.
255 */
256#if 0
257 printk("do_page_fault() #3: sending SIGBUS to %s for "
258 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
259 tsk->comm,
260 write ? "write access to" : "read access from",
261 field, address,
262 field, (unsigned long) regs->cp0_epc,
263 field, (unsigned long) regs->regs[31]);
264#endif
265 tsk->thread.cp0_badvaddr = address;
266 info.si_signo = SIGBUS;
267 info.si_errno = 0;
268 info.si_code = BUS_ADRERR;
269 info.si_addr = (void __user *) address;
270 force_sig_info(SIGBUS, &info, tsk);
271
272 return;
273#ifndef CONFIG_64BIT
274vmalloc_fault:
275 {
276 /*
277 * Synchronize this task's top level page-table
278 * with the 'reference' page table.
279 *
280 * Do _not_ use "tsk" here. We might be inside
281 * an interrupt in the middle of a task switch..
282 */
283 int offset = __pgd_offset(address);
284 pgd_t *pgd, *pgd_k;
285 pud_t *pud, *pud_k;
286 pmd_t *pmd, *pmd_k;
287 pte_t *pte_k;
288
289 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
290 pgd_k = init_mm.pgd + offset;
291
292 if (!pgd_present(*pgd_k))
293 goto no_context;
294 set_pgd(pgd, *pgd_k);
295
296 pud = pud_offset(pgd, address);
297 pud_k = pud_offset(pgd_k, address);
298 if (!pud_present(*pud_k))
299 goto no_context;
300
301 pmd = pmd_offset(pud, address);
302 pmd_k = pmd_offset(pud_k, address);
303 if (!pmd_present(*pmd_k))
304 goto no_context;
305 set_pmd(pmd, *pmd_k);
306
307 pte_k = pte_offset_kernel(pmd_k, address);
308 if (!pte_present(*pte_k))
309 goto no_context;
310 return;
311 }
312#endif
313}