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 return;
141
142 if (unlikely(fault & VM_FAULT_ERROR)) {
143 if (fault & VM_FAULT_OOM)
144 goto out_of_memory;
145 else if (fault & VM_FAULT_SIGSEGV)
146 goto bad_area;
147 else if (fault & VM_FAULT_SIGBUS)
148 goto do_sigbus;
149 BUG();
150 }
151
152 if (fault & VM_FAULT_RETRY) {
153 flags |= FAULT_FLAG_TRIED;
154
155 /*
156 * No need to mmap_read_unlock(mm) as we would
157 * have already released it in __lock_page_or_retry
158 * in mm/filemap.c.
159 */
160
161 goto retry;
162 }
163
164 mmap_read_unlock(mm);
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 */
171 bad_area:
172 mmap_read_unlock(mm);
173
174 bad_area_nosemaphore:
175 /* User mode accesses just cause a SIGSEGV */
176 if (user_mode(regs)) {
177 if (unhandled_signal(current, SIGSEGV) && printk_ratelimit()) {
178 pr_info("%s: unhandled page fault (%d) at 0x%08lx, "
179 "cause %ld\n", current->comm, SIGSEGV, address, cause);
180 show_regs(regs);
181 }
182 _exception(SIGSEGV, regs, code, address);
183 return;
184 }
185
186 no_context:
187 /* Are we prepared to handle this kernel fault? */
188 if (fixup_exception(regs))
189 return;
190
191 /*
192 * Oops. The kernel tried to access some bad page. We'll have to
193 * terminate things with extreme prejudice.
194 */
195 bust_spinlocks(1);
196
197 pr_alert("Unable to handle kernel %s at virtual address %08lx",
198 address < PAGE_SIZE ? "NULL pointer dereference" :
199 "paging request", address);
200 pr_alert("ea = %08lx, ra = %08lx, cause = %ld\n", regs->ea, regs->ra,
201 cause);
202 panic("Oops");
203 return;
204
205 /*
206 * We ran out of memory, or some other thing happened to us that made
207 * us unable to handle the page fault gracefully.
208 */
209 out_of_memory:
210 mmap_read_unlock(mm);
211 if (!user_mode(regs))
212 goto no_context;
213 pagefault_out_of_memory();
214 return;
215
216 do_sigbus:
217 mmap_read_unlock(mm);
218
219 /* Kernel mode? Handle exceptions or die */
220 if (!user_mode(regs))
221 goto no_context;
222
223 _exception(SIGBUS, regs, BUS_ADRERR, address);
224 return;
225
226 vmalloc_fault:
227 {
228 /*
229 * Synchronize this task's top level page-table
230 * with the 'reference' page table.
231 *
232 * Do _not_ use "tsk" here. We might be inside
233 * an interrupt in the middle of a task switch..
234 */
235 int offset = pgd_index(address);
236 pgd_t *pgd, *pgd_k;
237 p4d_t *p4d, *p4d_k;
238 pud_t *pud, *pud_k;
239 pmd_t *pmd, *pmd_k;
240 pte_t *pte_k;
241
242 pgd = pgd_current + offset;
243 pgd_k = init_mm.pgd + offset;
244
245 if (!pgd_present(*pgd_k))
246 goto no_context;
247 set_pgd(pgd, *pgd_k);
248
249 p4d = p4d_offset(pgd, address);
250 p4d_k = p4d_offset(pgd_k, address);
251 if (!p4d_present(*p4d_k))
252 goto no_context;
253 pud = pud_offset(p4d, address);
254 pud_k = pud_offset(p4d_k, address);
255 if (!pud_present(*pud_k))
256 goto no_context;
257 pmd = pmd_offset(pud, address);
258 pmd_k = pmd_offset(pud_k, address);
259 if (!pmd_present(*pmd_k))
260 goto no_context;
261 set_pmd(pmd, *pmd_k);
262
263 pte_k = pte_offset_kernel(pmd_k, address);
264 if (!pte_present(*pte_k))
265 goto no_context;
266
267 flush_tlb_kernel_page(address);
268 return;
269 }
270 }
271