1 /*
2 * Copyright (C) 2009 Wind River Systems Inc
3 * Implemented by fredrik.markstrom@gmail.com and ivarholmqvist@gmail.com
4 *
5 * based on arch/mips/mm/fault.c which is:
6 *
7 * Copyright (C) 1995-2000 Ralf Baechle
8 *
9 * This file is subject to the terms and conditions of the GNU General Public
10 * License. See the file "COPYING" in the main directory of this archive
11 * for more details.
12 */
13
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/interrupt.h>
18 #include <linux/kernel.h>
19 #include <linux/errno.h>
20 #include <linux/string.h>
21 #include <linux/types.h>
22 #include <linux/ptrace.h>
23 #include <linux/mman.h>
24 #include <linux/mm.h>
25 #include <linux/extable.h>
26 #include <linux/uaccess.h>
27 #include <linux/perf_event.h>
28
29 #include <asm/mmu_context.h>
30 #include <asm/traps.h>
31
32 #define EXC_SUPERV_INSN_ACCESS 9 /* Supervisor only instruction address */
33 #define EXC_SUPERV_DATA_ACCESS 11 /* Supervisor only data address */
34 #define EXC_X_PROTECTION_FAULT 13 /* TLB permission violation (x) */
35 #define EXC_R_PROTECTION_FAULT 14 /* TLB permission violation (r) */
36 #define EXC_W_PROTECTION_FAULT 15 /* TLB permission violation (w) */
37
38 /*
39 * This routine handles page faults. It determines the address,
40 * and the problem, and then passes it off to one of the appropriate
41 * routines.
42 */
do_page_fault(struct pt_regs * regs,unsigned long cause,unsigned long address)43 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long cause,
44 unsigned long address)
45 {
46 struct vm_area_struct *vma = NULL;
47 struct task_struct *tsk = current;
48 struct mm_struct *mm = tsk->mm;
49 int code = SEGV_MAPERR;
50 vm_fault_t fault;
51 unsigned int flags = FAULT_FLAG_DEFAULT;
52
53 cause >>= 2;
54
55 /* Restart the instruction */
56 regs->ea -= 4;
57
58 /*
59 * We fault-in kernel-space virtual memory on-demand. The
60 * 'reference' page table is init_mm.pgd.
61 *
62 * NOTE! We MUST NOT take any locks for this case. We may
63 * be in an interrupt or a critical region, and should
64 * only copy the information from the master page table,
65 * nothing more.
66 */
67 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END)) {
68 if (user_mode(regs))
69 goto bad_area_nosemaphore;
70 else
71 goto vmalloc_fault;
72 }
73
74 if (unlikely(address >= TASK_SIZE))
75 goto bad_area_nosemaphore;
76
77 /*
78 * If we're in an interrupt or have no user
79 * context, we must not take the fault..
80 */
81 if (faulthandler_disabled() || !mm)
82 goto bad_area_nosemaphore;
83
84 if (user_mode(regs))
85 flags |= FAULT_FLAG_USER;
86
87 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
88
89 if (!mmap_read_trylock(mm)) {
90 if (!user_mode(regs) && !search_exception_tables(regs->ea))
91 goto bad_area_nosemaphore;
92 retry:
93 mmap_read_lock(mm);
94 }
95
96 vma = find_vma(mm, address);
97 if (!vma)
98 goto bad_area;
99 if (vma->vm_start <= address)
100 goto good_area;
101 if (!(vma->vm_flags & VM_GROWSDOWN))
102 goto bad_area;
103 if (expand_stack(vma, address))
104 goto bad_area;
105 /*
106 * Ok, we have a good vm_area for this memory access, so
107 * we can handle it..
108 */
109 good_area:
110 code = SEGV_ACCERR;
111
112 switch (cause) {
113 case EXC_SUPERV_INSN_ACCESS:
114 goto bad_area;
115 case EXC_SUPERV_DATA_ACCESS:
116 goto bad_area;
117 case EXC_X_PROTECTION_FAULT:
118 if (!(vma->vm_flags & VM_EXEC))
119 goto bad_area;
120 break;
121 case EXC_R_PROTECTION_FAULT:
122 if (!(vma->vm_flags & VM_READ))
123 goto bad_area;
124 break;
125 case EXC_W_PROTECTION_FAULT:
126 if (!(vma->vm_flags & VM_WRITE))
127 goto bad_area;
128 flags = FAULT_FLAG_WRITE;
129 break;
130 }
131
132 /*
133 * If for any reason at all we couldn't handle the fault,
134 * make sure we exit gracefully rather than endlessly redo
135 * the fault.
136 */
137 fault = handle_mm_fault(vma, address, flags, regs);
138
139 if (fault_signal_pending(fault, regs)) {
140 if (!user_mode(regs))
141 goto no_context;
142 return;
143 }
144
145 /* The fault is fully completed (including releasing mmap lock) */
146 if (fault & VM_FAULT_COMPLETED)
147 return;
148
149 if (unlikely(fault & VM_FAULT_ERROR)) {
150 if (fault & VM_FAULT_OOM)
151 goto out_of_memory;
152 else if (fault & VM_FAULT_SIGSEGV)
153 goto bad_area;
154 else if (fault & VM_FAULT_SIGBUS)
155 goto do_sigbus;
156 BUG();
157 }
158
159 if (fault & VM_FAULT_RETRY) {
160 flags |= FAULT_FLAG_TRIED;
161
162 /*
163 * No need to mmap_read_unlock(mm) as we would
164 * have already released it in __lock_page_or_retry
165 * in mm/filemap.c.
166 */
167
168 goto retry;
169 }
170
171 mmap_read_unlock(mm);
172 return;
173
174 /*
175 * Something tried to access memory that isn't in our memory map..
176 * Fix it, but check if it's kernel or user first..
177 */
178 bad_area:
179 mmap_read_unlock(mm);
180
181 bad_area_nosemaphore:
182 /* User mode accesses just cause a SIGSEGV */
183 if (user_mode(regs)) {
184 if (unhandled_signal(current, SIGSEGV) && printk_ratelimit()) {
185 pr_info("%s: unhandled page fault (%d) at 0x%08lx, "
186 "cause %ld\n", current->comm, SIGSEGV, address, cause);
187 show_regs(regs);
188 }
189 _exception(SIGSEGV, regs, code, address);
190 return;
191 }
192
193 no_context:
194 /* Are we prepared to handle this kernel fault? */
195 if (fixup_exception(regs))
196 return;
197
198 /*
199 * Oops. The kernel tried to access some bad page. We'll have to
200 * terminate things with extreme prejudice.
201 */
202 bust_spinlocks(1);
203
204 pr_alert("Unable to handle kernel %s at virtual address %08lx",
205 address < PAGE_SIZE ? "NULL pointer dereference" :
206 "paging request", address);
207 pr_alert("ea = %08lx, ra = %08lx, cause = %ld\n", regs->ea, regs->ra,
208 cause);
209 panic("Oops");
210 return;
211
212 /*
213 * We ran out of memory, or some other thing happened to us that made
214 * us unable to handle the page fault gracefully.
215 */
216 out_of_memory:
217 mmap_read_unlock(mm);
218 if (!user_mode(regs))
219 goto no_context;
220 pagefault_out_of_memory();
221 return;
222
223 do_sigbus:
224 mmap_read_unlock(mm);
225
226 /* Kernel mode? Handle exceptions or die */
227 if (!user_mode(regs))
228 goto no_context;
229
230 _exception(SIGBUS, regs, BUS_ADRERR, address);
231 return;
232
233 vmalloc_fault:
234 {
235 /*
236 * Synchronize this task's top level page-table
237 * with the 'reference' page table.
238 *
239 * Do _not_ use "tsk" here. We might be inside
240 * an interrupt in the middle of a task switch..
241 */
242 int offset = pgd_index(address);
243 pgd_t *pgd, *pgd_k;
244 p4d_t *p4d, *p4d_k;
245 pud_t *pud, *pud_k;
246 pmd_t *pmd, *pmd_k;
247 pte_t *pte_k;
248
249 pgd = pgd_current + offset;
250 pgd_k = init_mm.pgd + offset;
251
252 if (!pgd_present(*pgd_k))
253 goto no_context;
254 set_pgd(pgd, *pgd_k);
255
256 p4d = p4d_offset(pgd, address);
257 p4d_k = p4d_offset(pgd_k, address);
258 if (!p4d_present(*p4d_k))
259 goto no_context;
260 pud = pud_offset(p4d, address);
261 pud_k = pud_offset(p4d_k, address);
262 if (!pud_present(*pud_k))
263 goto no_context;
264 pmd = pmd_offset(pud, address);
265 pmd_k = pmd_offset(pud_k, address);
266 if (!pmd_present(*pmd_k))
267 goto no_context;
268 set_pmd(pmd, *pmd_k);
269
270 pte_k = pte_offset_kernel(pmd_k, address);
271 if (!pte_present(*pte_k))
272 goto no_context;
273
274 flush_tlb_kernel_page(address);
275 return;
276 }
277 }
278