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/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/system.h>
26#include <asm/uaccess.h>
27#include <asm/ptrace.h>
28#include <asm/highmem.h> /* For VMALLOC_END */
29#include <linux/kdebug.h>
30
31/*
32 * This routine handles page faults. It determines the address,
33 * and the problem, and then passes it off to one of the appropriate
34 * routines.
35 */
36asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, unsigned long write,
37 unsigned long address)
38{
39 struct vm_area_struct * vma = NULL;
40 struct task_struct *tsk = current;
41 struct mm_struct *mm = tsk->mm;
42 const int field = sizeof(unsigned long) * 2;
43 siginfo_t info;
44 int fault;
45
46#if 0
47 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
48 current->comm, current->pid, field, address, write,
49 field, regs->cp0_epc);
50#endif
51
52#ifdef CONFIG_KPROBES
53 /*
54 * This is to notify the fault handler of the kprobes. The
55 * exception code is redundant as it is also carried in REGS,
56 * but we pass it anyhow.
57 */
58 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
59 (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
60 return;
61#endif
62
63 info.si_code = SEGV_MAPERR;
64
65 /*
66 * We fault-in kernel-space virtual memory on-demand. The
67 * 'reference' page table is init_mm.pgd.
68 *
69 * NOTE! We MUST NOT take any locks for this case. We may
70 * be in an interrupt or a critical region, and should
71 * only copy the information from the master page table,
72 * nothing more.
73 */
74#ifdef CONFIG_64BIT
75# define VMALLOC_FAULT_TARGET no_context
76#else
77# define VMALLOC_FAULT_TARGET vmalloc_fault
78#endif
79
80 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
81 goto VMALLOC_FAULT_TARGET;
82#ifdef MODULE_START
83 if (unlikely(address >= MODULE_START && address < MODULE_END))
84 goto VMALLOC_FAULT_TARGET;
85#endif
86
87 /*
88 * If we're in an interrupt or have no user
89 * context, we must not take the fault..
90 */
91 if (in_atomic() || !mm)
92 goto bad_area_nosemaphore;
93
94 down_read(&mm->mmap_sem);
95 vma = find_vma(mm, address);
96 if (!vma)
97 goto bad_area;
98 if (vma->vm_start <= address)
99 goto good_area;
100 if (!(vma->vm_flags & VM_GROWSDOWN))
101 goto bad_area;
102 if (expand_stack(vma, address))
103 goto bad_area;
104/*
105 * Ok, we have a good vm_area for this memory access, so
106 * we can handle it..
107 */
108good_area:
109 info.si_code = SEGV_ACCERR;
110
111 if (write) {
112 if (!(vma->vm_flags & VM_WRITE))
113 goto bad_area;
114 } else {
115 if (kernel_uses_smartmips_rixi) {
116 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
117#if 0
118 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
119 raw_smp_processor_id(),
120 current->comm, current->pid,
121 field, address, write,
122 field, regs->cp0_epc);
123#endif
124 goto bad_area;
125 }
126 if (!(vma->vm_flags & VM_READ)) {
127#if 0
128 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
129 raw_smp_processor_id(),
130 current->comm, current->pid,
131 field, address, write,
132 field, regs->cp0_epc);
133#endif
134 goto bad_area;
135 }
136 } else {
137 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
138 goto bad_area;
139 }
140 }
141
142 /*
143 * If for any reason at all we couldn't handle the fault,
144 * make sure we exit gracefully rather than endlessly redo
145 * the fault.
146 */
147 fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0);
148 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
149 if (unlikely(fault & VM_FAULT_ERROR)) {
150 if (fault & VM_FAULT_OOM)
151 goto out_of_memory;
152 else if (fault & VM_FAULT_SIGBUS)
153 goto do_sigbus;
154 BUG();
155 }
156 if (fault & VM_FAULT_MAJOR) {
157 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
158 tsk->maj_flt++;
159 } else {
160 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
161 tsk->min_flt++;
162 }
163
164 up_read(&mm->mmap_sem);
165 return;
166
167/*
168 * Something tried to access memory that isn't in our memory map..
169 * Fix it, but check if it's kernel or user first..
170 */
171bad_area:
172 up_read(&mm->mmap_sem);
173
174bad_area_nosemaphore:
175 /* User mode accesses just cause a SIGSEGV */
176 if (user_mode(regs)) {
177 tsk->thread.cp0_badvaddr = address;
178 tsk->thread.error_code = write;
179#if 0
180 printk("do_page_fault() #2: sending SIGSEGV to %s for "
181 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
182 tsk->comm,
183 write ? "write access to" : "read access from",
184 field, address,
185 field, (unsigned long) regs->cp0_epc,
186 field, (unsigned long) regs->regs[31]);
187#endif
188 info.si_signo = SIGSEGV;
189 info.si_errno = 0;
190 /* info.si_code has been set above */
191 info.si_addr = (void __user *) address;
192 force_sig_info(SIGSEGV, &info, tsk);
193 return;
194 }
195
196no_context:
197 /* Are we prepared to handle this kernel fault? */
198 if (fixup_exception(regs)) {
199 current->thread.cp0_baduaddr = address;
200 return;
201 }
202
203 /*
204 * Oops. The kernel tried to access some bad page. We'll have to
205 * terminate things with extreme prejudice.
206 */
207 bust_spinlocks(1);
208
209 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
210 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
211 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
212 field, regs->regs[31]);
213 die("Oops", regs);
214
215out_of_memory:
216 /*
217 * We ran out of memory, call the OOM killer, and return the userspace
218 * (which will retry the fault, or kill us if we got oom-killed).
219 */
220 up_read(&mm->mmap_sem);
221 pagefault_out_of_memory();
222 return;
223
224do_sigbus:
225 up_read(&mm->mmap_sem);
226
227 /* Kernel mode? Handle exceptions or die */
228 if (!user_mode(regs))
229 goto no_context;
230 else
231 /*
232 * Send a sigbus, regardless of whether we were in kernel
233 * or user mode.
234 */
235#if 0
236 printk("do_page_fault() #3: sending SIGBUS to %s for "
237 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
238 tsk->comm,
239 write ? "write access to" : "read access from",
240 field, address,
241 field, (unsigned long) regs->cp0_epc,
242 field, (unsigned long) regs->regs[31]);
243#endif
244 tsk->thread.cp0_badvaddr = address;
245 info.si_signo = SIGBUS;
246 info.si_errno = 0;
247 info.si_code = BUS_ADRERR;
248 info.si_addr = (void __user *) address;
249 force_sig_info(SIGBUS, &info, tsk);
250
251 return;
252#ifndef CONFIG_64BIT
253vmalloc_fault:
254 {
255 /*
256 * Synchronize this task's top level page-table
257 * with the 'reference' page table.
258 *
259 * Do _not_ use "tsk" here. We might be inside
260 * an interrupt in the middle of a task switch..
261 */
262 int offset = __pgd_offset(address);
263 pgd_t *pgd, *pgd_k;
264 pud_t *pud, *pud_k;
265 pmd_t *pmd, *pmd_k;
266 pte_t *pte_k;
267
268 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
269 pgd_k = init_mm.pgd + offset;
270
271 if (!pgd_present(*pgd_k))
272 goto no_context;
273 set_pgd(pgd, *pgd_k);
274
275 pud = pud_offset(pgd, address);
276 pud_k = pud_offset(pgd_k, address);
277 if (!pud_present(*pud_k))
278 goto no_context;
279
280 pmd = pmd_offset(pud, address);
281 pmd_k = pmd_offset(pud_k, address);
282 if (!pmd_present(*pmd_k))
283 goto no_context;
284 set_pmd(pmd, *pmd_k);
285
286 pte_k = pte_offset_kernel(pmd_k, address);
287 if (!pte_present(*pte_k))
288 goto no_context;
289 return;
290 }
291#endif
292}
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 __kprobes __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 siginfo_t info;
46 int fault;
47 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
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 info.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;
99retry:
100 down_read(&mm->mmap_sem);
101 vma = find_vma(mm, address);
102 if (!vma)
103 goto bad_area;
104 if (vma->vm_start <= address)
105 goto good_area;
106 if (!(vma->vm_flags & VM_GROWSDOWN))
107 goto bad_area;
108 if (expand_stack(vma, address))
109 goto bad_area;
110/*
111 * Ok, we have a good vm_area for this memory access, so
112 * we can handle it..
113 */
114good_area:
115 info.si_code = SEGV_ACCERR;
116
117 if (write) {
118 if (!(vma->vm_flags & VM_WRITE))
119 goto bad_area;
120 flags |= FAULT_FLAG_WRITE;
121 } else {
122 if (cpu_has_rixi) {
123 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
124#if 0
125 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
126 raw_smp_processor_id(),
127 current->comm, current->pid,
128 field, address, write,
129 field, regs->cp0_epc);
130#endif
131 goto bad_area;
132 }
133 if (!(vma->vm_flags & VM_READ) &&
134 exception_epc(regs) != address) {
135#if 0
136 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
137 raw_smp_processor_id(),
138 current->comm, current->pid,
139 field, address, write,
140 field, regs->cp0_epc);
141#endif
142 goto bad_area;
143 }
144 } else {
145 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
146 goto bad_area;
147 }
148 }
149
150 /*
151 * If for any reason at all we couldn't handle the fault,
152 * make sure we exit gracefully rather than endlessly redo
153 * the fault.
154 */
155 fault = handle_mm_fault(vma, address, flags);
156
157 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
158 return;
159
160 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
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 if (flags & FAULT_FLAG_ALLOW_RETRY) {
171 if (fault & VM_FAULT_MAJOR) {
172 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
173 regs, address);
174 tsk->maj_flt++;
175 } else {
176 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
177 regs, address);
178 tsk->min_flt++;
179 }
180 if (fault & VM_FAULT_RETRY) {
181 flags &= ~FAULT_FLAG_ALLOW_RETRY;
182 flags |= FAULT_FLAG_TRIED;
183
184 /*
185 * No need to up_read(&mm->mmap_sem) as we would
186 * have already released it in __lock_page_or_retry
187 * in mm/filemap.c.
188 */
189
190 goto retry;
191 }
192 }
193
194 up_read(&mm->mmap_sem);
195 return;
196
197/*
198 * Something tried to access memory that isn't in our memory map..
199 * Fix it, but check if it's kernel or user first..
200 */
201bad_area:
202 up_read(&mm->mmap_sem);
203
204bad_area_nosemaphore:
205 /* User mode accesses just cause a SIGSEGV */
206 if (user_mode(regs)) {
207 tsk->thread.cp0_badvaddr = address;
208 tsk->thread.error_code = write;
209 if (show_unhandled_signals &&
210 unhandled_signal(tsk, SIGSEGV) &&
211 __ratelimit(&ratelimit_state)) {
212 pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n",
213 tsk->comm,
214 write ? "write access to" : "read access from",
215 field, address);
216 pr_info("epc = %0*lx in", field,
217 (unsigned long) regs->cp0_epc);
218 print_vma_addr(KERN_CONT " ", regs->cp0_epc);
219 pr_cont("\n");
220 pr_info("ra = %0*lx in", field,
221 (unsigned long) regs->regs[31]);
222 print_vma_addr(KERN_CONT " ", regs->regs[31]);
223 pr_cont("\n");
224 }
225 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
226 info.si_signo = SIGSEGV;
227 info.si_errno = 0;
228 /* info.si_code has been set above */
229 info.si_addr = (void __user *) address;
230 force_sig_info(SIGSEGV, &info, tsk);
231 return;
232 }
233
234no_context:
235 /* Are we prepared to handle this kernel fault? */
236 if (fixup_exception(regs)) {
237 current->thread.cp0_baduaddr = address;
238 return;
239 }
240
241 /*
242 * Oops. The kernel tried to access some bad page. We'll have to
243 * terminate things with extreme prejudice.
244 */
245 bust_spinlocks(1);
246
247 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
248 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
249 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
250 field, regs->regs[31]);
251 die("Oops", regs);
252
253out_of_memory:
254 /*
255 * We ran out of memory, call the OOM killer, and return the userspace
256 * (which will retry the fault, or kill us if we got oom-killed).
257 */
258 up_read(&mm->mmap_sem);
259 if (!user_mode(regs))
260 goto no_context;
261 pagefault_out_of_memory();
262 return;
263
264do_sigbus:
265 up_read(&mm->mmap_sem);
266
267 /* Kernel mode? Handle exceptions or die */
268 if (!user_mode(regs))
269 goto no_context;
270
271 /*
272 * Send a sigbus, regardless of whether we were in kernel
273 * or user mode.
274 */
275#if 0
276 printk("do_page_fault() #3: sending SIGBUS to %s for "
277 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
278 tsk->comm,
279 write ? "write access to" : "read access from",
280 field, address,
281 field, (unsigned long) regs->cp0_epc,
282 field, (unsigned long) regs->regs[31]);
283#endif
284 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
285 tsk->thread.cp0_badvaddr = address;
286 info.si_signo = SIGBUS;
287 info.si_errno = 0;
288 info.si_code = BUS_ADRERR;
289 info.si_addr = (void __user *) address;
290 force_sig_info(SIGBUS, &info, tsk);
291
292 return;
293#ifndef CONFIG_64BIT
294vmalloc_fault:
295 {
296 /*
297 * Synchronize this task's top level page-table
298 * with the 'reference' page table.
299 *
300 * Do _not_ use "tsk" here. We might be inside
301 * an interrupt in the middle of a task switch..
302 */
303 int offset = __pgd_offset(address);
304 pgd_t *pgd, *pgd_k;
305 pud_t *pud, *pud_k;
306 pmd_t *pmd, *pmd_k;
307 pte_t *pte_k;
308
309 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
310 pgd_k = init_mm.pgd + offset;
311
312 if (!pgd_present(*pgd_k))
313 goto no_context;
314 set_pgd(pgd, *pgd_k);
315
316 pud = pud_offset(pgd, address);
317 pud_k = pud_offset(pgd_k, address);
318 if (!pud_present(*pud_k))
319 goto no_context;
320
321 pmd = pmd_offset(pud, address);
322 pmd_k = pmd_offset(pud_k, address);
323 if (!pmd_present(*pmd_k))
324 goto no_context;
325 set_pmd(pmd, *pmd_k);
326
327 pte_k = pte_offset_kernel(pmd_k, address);
328 if (!pte_present(*pte_k))
329 goto no_context;
330 return;
331 }
332#endif
333}
334
335asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
336 unsigned long write, unsigned long address)
337{
338 enum ctx_state prev_state;
339
340 prev_state = exception_enter();
341 __do_page_fault(regs, write, address);
342 exception_exit(prev_state);
343}