Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
028c1f68 | 2 | * arch/cris/mm/fault.c |
1da177e4 | 3 | * |
028c1f68 | 4 | * Copyright (C) 2000-2010 Axis Communications AB |
1da177e4 LT |
5 | */ |
6 | ||
7 | #include <linux/mm.h> | |
8 | #include <linux/interrupt.h> | |
9 | #include <linux/module.h> | |
b4e8a181 | 10 | #include <linux/wait.h> |
1da177e4 | 11 | #include <asm/uaccess.h> |
b1a154db | 12 | #include <arch/system.h> |
1da177e4 LT |
13 | |
14 | extern int find_fixup_code(struct pt_regs *); | |
15 | extern void die_if_kernel(const char *, struct pt_regs *, long); | |
2d495ebc | 16 | extern void show_registers(struct pt_regs *regs); |
1da177e4 LT |
17 | |
18 | /* debug of low-level TLB reload */ | |
19 | #undef DEBUG | |
20 | ||
21 | #ifdef DEBUG | |
22 | #define D(x) x | |
23 | #else | |
24 | #define D(x) | |
25 | #endif | |
26 | ||
27 | /* debug of higher-level faults */ | |
28 | #define DPG(x) | |
29 | ||
30 | /* current active page directory */ | |
31 | ||
fe87f94f | 32 | DEFINE_PER_CPU(pgd_t *, current_pgd); |
4f18cfbf | 33 | unsigned long cris_signal_return_page; |
1da177e4 LT |
34 | |
35 | /* | |
36 | * This routine handles page faults. It determines the address, | |
37 | * and the problem, and then passes it off to one of the appropriate | |
38 | * routines. | |
39 | * | |
40 | * Notice that the address we're given is aligned to the page the fault | |
41 | * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete | |
42 | * address. | |
43 | * | |
44 | * error_code: | |
3e1fdc4e JN |
45 | * bit 0 == 0 means no page found, 1 means protection fault |
46 | * bit 1 == 0 means read, 1 means write | |
1da177e4 LT |
47 | * |
48 | * If this routine detects a bad access, it returns 1, otherwise it | |
49 | * returns 0. | |
50 | */ | |
51 | ||
52 | asmlinkage void | |
53 | do_page_fault(unsigned long address, struct pt_regs *regs, | |
54 | int protection, int writeaccess) | |
55 | { | |
56 | struct task_struct *tsk; | |
57 | struct mm_struct *mm; | |
58 | struct vm_area_struct * vma; | |
59 | siginfo_t info; | |
83c54070 | 60 | int fault; |
759496ba | 61 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 62 | |
3e1fdc4e JN |
63 | D(printk(KERN_DEBUG |
64 | "Page fault for %lX on %X at %lX, prot %d write %d\n", | |
65 | address, smp_processor_id(), instruction_pointer(regs), | |
66 | protection, writeaccess)); | |
1da177e4 LT |
67 | |
68 | tsk = current; | |
69 | ||
70 | /* | |
71 | * We fault-in kernel-space virtual memory on-demand. The | |
72 | * 'reference' page table is init_mm.pgd. | |
73 | * | |
74 | * NOTE! We MUST NOT take any locks for this case. We may | |
75 | * be in an interrupt or a critical region, and should | |
76 | * only copy the information from the master page table, | |
77 | * nothing more. | |
78 | * | |
79 | * NOTE2: This is done so that, when updating the vmalloc | |
80 | * mappings we don't have to walk all processes pgdirs and | |
81 | * add the high mappings all at once. Instead we do it as they | |
82 | * are used. However vmalloc'ed page entries have the PAGE_GLOBAL | |
83 | * bit set so sometimes the TLB can use a lingering entry. | |
84 | * | |
85 | * This verifies that the fault happens in kernel space | |
86 | * and that the fault was not a protection error (error_code & 1). | |
87 | */ | |
88 | ||
89 | if (address >= VMALLOC_START && | |
90 | !protection && | |
91 | !user_mode(regs)) | |
92 | goto vmalloc_fault; | |
93 | ||
4f18cfbf MS |
94 | /* When stack execution is not allowed we store the signal |
95 | * trampolines in the reserved cris_signal_return_page. | |
96 | * Handle this in the exact same way as vmalloc (we know | |
97 | * that the mapping is there and is valid so no need to | |
98 | * call handle_mm_fault). | |
99 | */ | |
100 | if (cris_signal_return_page && | |
101 | address == cris_signal_return_page && | |
102 | !protection && user_mode(regs)) | |
103 | goto vmalloc_fault; | |
104 | ||
1da177e4 | 105 | /* we can and should enable interrupts at this point */ |
4f18cfbf | 106 | local_irq_enable(); |
1da177e4 LT |
107 | |
108 | mm = tsk->mm; | |
109 | info.si_code = SEGV_MAPERR; | |
110 | ||
111 | /* | |
028c1f68 JN |
112 | * If we're in an interrupt or "atomic" operation or have no |
113 | * user context, we must not take the fault. | |
1da177e4 LT |
114 | */ |
115 | ||
028c1f68 | 116 | if (in_atomic() || !mm) |
1da177e4 LT |
117 | goto no_context; |
118 | ||
759496ba JW |
119 | if (user_mode(regs)) |
120 | flags |= FAULT_FLAG_USER; | |
4d5914d6 | 121 | retry: |
1da177e4 LT |
122 | down_read(&mm->mmap_sem); |
123 | vma = find_vma(mm, address); | |
124 | if (!vma) | |
125 | goto bad_area; | |
126 | if (vma->vm_start <= address) | |
127 | goto good_area; | |
128 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
129 | goto bad_area; | |
130 | if (user_mode(regs)) { | |
131 | /* | |
132 | * accessing the stack below usp is always a bug. | |
133 | * we get page-aligned addresses so we can only check | |
134 | * if we're within a page from usp, but that might be | |
135 | * enough to catch brutal errors at least. | |
136 | */ | |
137 | if (address + PAGE_SIZE < rdusp()) | |
138 | goto bad_area; | |
139 | } | |
140 | if (expand_stack(vma, address)) | |
141 | goto bad_area; | |
142 | ||
143 | /* | |
144 | * Ok, we have a good vm_area for this memory access, so | |
145 | * we can handle it.. | |
146 | */ | |
147 | ||
148 | good_area: | |
149 | info.si_code = SEGV_ACCERR; | |
150 | ||
151 | /* first do some preliminary protection checks */ | |
152 | ||
4f18cfbf MS |
153 | if (writeaccess == 2){ |
154 | if (!(vma->vm_flags & VM_EXEC)) | |
155 | goto bad_area; | |
156 | } else if (writeaccess == 1) { | |
1da177e4 LT |
157 | if (!(vma->vm_flags & VM_WRITE)) |
158 | goto bad_area; | |
759496ba | 159 | flags |= FAULT_FLAG_WRITE; |
1da177e4 LT |
160 | } else { |
161 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) | |
162 | goto bad_area; | |
163 | } | |
164 | ||
165 | /* | |
166 | * If for any reason at all we couldn't handle the fault, | |
167 | * make sure we exit gracefully rather than endlessly redo | |
168 | * the fault. | |
169 | */ | |
170 | ||
4d5914d6 KC |
171 | fault = handle_mm_fault(mm, vma, address, flags); |
172 | ||
173 | if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) | |
174 | return; | |
175 | ||
83c54070 NP |
176 | if (unlikely(fault & VM_FAULT_ERROR)) { |
177 | if (fault & VM_FAULT_OOM) | |
178 | goto out_of_memory; | |
33692f27 LT |
179 | else if (fault & VM_FAULT_SIGSEGV) |
180 | goto bad_area; | |
83c54070 NP |
181 | else if (fault & VM_FAULT_SIGBUS) |
182 | goto do_sigbus; | |
183 | BUG(); | |
1da177e4 | 184 | } |
4d5914d6 KC |
185 | |
186 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
187 | if (fault & VM_FAULT_MAJOR) | |
188 | tsk->maj_flt++; | |
189 | else | |
190 | tsk->min_flt++; | |
191 | if (fault & VM_FAULT_RETRY) { | |
192 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
45cac65b | 193 | flags |= FAULT_FLAG_TRIED; |
4d5914d6 KC |
194 | |
195 | /* | |
196 | * No need to up_read(&mm->mmap_sem) as we would | |
197 | * have already released it in __lock_page_or_retry | |
198 | * in mm/filemap.c. | |
199 | */ | |
200 | ||
201 | goto retry; | |
202 | } | |
203 | } | |
1da177e4 LT |
204 | |
205 | up_read(&mm->mmap_sem); | |
206 | return; | |
207 | ||
208 | /* | |
209 | * Something tried to access memory that isn't in our memory map.. | |
210 | * Fix it, but check if it's kernel or user first.. | |
211 | */ | |
212 | ||
213 | bad_area: | |
214 | up_read(&mm->mmap_sem); | |
215 | ||
216 | bad_area_nosemaphore: | |
217 | DPG(show_registers(regs)); | |
218 | ||
219 | /* User mode accesses just cause a SIGSEGV */ | |
220 | ||
221 | if (user_mode(regs)) { | |
b4e8a181 JN |
222 | printk(KERN_NOTICE "%s (pid %d) segfaults for page " |
223 | "address %08lx at pc %08lx\n", | |
224 | tsk->comm, tsk->pid, | |
225 | address, instruction_pointer(regs)); | |
2d495ebc JN |
226 | |
227 | /* With DPG on, we've already dumped registers above. */ | |
228 | DPG(if (0)) | |
229 | show_registers(regs); | |
230 | ||
b4e8a181 JN |
231 | #ifdef CONFIG_NO_SEGFAULT_TERMINATION |
232 | DECLARE_WAIT_QUEUE_HEAD(wq); | |
233 | wait_event_interruptible(wq, 0 == 1); | |
234 | #else | |
1da177e4 LT |
235 | info.si_signo = SIGSEGV; |
236 | info.si_errno = 0; | |
237 | /* info.si_code has been set above */ | |
238 | info.si_addr = (void *)address; | |
239 | force_sig_info(SIGSEGV, &info, tsk); | |
b4e8a181 | 240 | #endif |
1da177e4 LT |
241 | return; |
242 | } | |
243 | ||
244 | no_context: | |
245 | ||
246 | /* Are we prepared to handle this kernel fault? | |
247 | * | |
3e1fdc4e | 248 | * (The kernel has valid exception-points in the source |
af901ca1 | 249 | * when it accesses user-memory. When it fails in one |
1da177e4 LT |
250 | * of those points, we find it in a table and do a jump |
251 | * to some fixup code that loads an appropriate error | |
252 | * code) | |
253 | */ | |
254 | ||
255 | if (find_fixup_code(regs)) | |
256 | return; | |
257 | ||
258 | /* | |
259 | * Oops. The kernel tried to access some bad page. We'll have to | |
260 | * terminate things with extreme prejudice. | |
261 | */ | |
262 | ||
3e1fdc4e JN |
263 | if (!oops_in_progress) { |
264 | oops_in_progress = 1; | |
265 | if ((unsigned long) (address) < PAGE_SIZE) | |
266 | printk(KERN_ALERT "Unable to handle kernel NULL " | |
267 | "pointer dereference"); | |
268 | else | |
269 | printk(KERN_ALERT "Unable to handle kernel access" | |
270 | " at virtual address %08lx\n", address); | |
271 | ||
272 | die_if_kernel("Oops", regs, (writeaccess << 1) | protection); | |
273 | oops_in_progress = 0; | |
274 | } | |
1da177e4 LT |
275 | |
276 | do_exit(SIGKILL); | |
277 | ||
278 | /* | |
279 | * We ran out of memory, or some other thing happened to us that made | |
280 | * us unable to handle the page fault gracefully. | |
281 | */ | |
282 | ||
283 | out_of_memory: | |
284 | up_read(&mm->mmap_sem); | |
3648bdf7 JN |
285 | if (!user_mode(regs)) |
286 | goto no_context; | |
287 | pagefault_out_of_memory(); | |
288 | return; | |
1da177e4 LT |
289 | |
290 | do_sigbus: | |
291 | up_read(&mm->mmap_sem); | |
292 | ||
293 | /* | |
294 | * Send a sigbus, regardless of whether we were in kernel | |
295 | * or user mode. | |
296 | */ | |
297 | info.si_signo = SIGBUS; | |
298 | info.si_errno = 0; | |
299 | info.si_code = BUS_ADRERR; | |
300 | info.si_addr = (void *)address; | |
301 | force_sig_info(SIGBUS, &info, tsk); | |
302 | ||
303 | /* Kernel mode? Handle exceptions or die */ | |
304 | if (!user_mode(regs)) | |
305 | goto no_context; | |
306 | return; | |
307 | ||
308 | vmalloc_fault: | |
309 | { | |
310 | /* | |
311 | * Synchronize this task's top level page-table | |
312 | * with the 'reference' page table. | |
313 | * | |
314 | * Use current_pgd instead of tsk->active_mm->pgd | |
315 | * since the latter might be unavailable if this | |
316 | * code is executed in a misfortunately run irq | |
317 | * (like inside schedule() between switch_mm and | |
318 | * switch_to...). | |
319 | */ | |
320 | ||
321 | int offset = pgd_index(address); | |
322 | pgd_t *pgd, *pgd_k; | |
4f18cfbf | 323 | pud_t *pud, *pud_k; |
1da177e4 LT |
324 | pmd_t *pmd, *pmd_k; |
325 | pte_t *pte_k; | |
326 | ||
4f18cfbf | 327 | pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset; |
1da177e4 LT |
328 | pgd_k = init_mm.pgd + offset; |
329 | ||
330 | /* Since we're two-level, we don't need to do both | |
331 | * set_pgd and set_pmd (they do the same thing). If | |
332 | * we go three-level at some point, do the right thing | |
3e1fdc4e JN |
333 | * with pgd_present and set_pgd here. |
334 | * | |
1da177e4 LT |
335 | * Also, since the vmalloc area is global, we don't |
336 | * need to copy individual PTE's, it is enough to | |
337 | * copy the pgd pointer into the pte page of the | |
338 | * root task. If that is there, we'll find our pte if | |
339 | * it exists. | |
340 | */ | |
341 | ||
4f18cfbf MS |
342 | pud = pud_offset(pgd, address); |
343 | pud_k = pud_offset(pgd_k, address); | |
344 | if (!pud_present(*pud_k)) | |
345 | goto no_context; | |
346 | ||
347 | pmd = pmd_offset(pud, address); | |
348 | pmd_k = pmd_offset(pud_k, address); | |
1da177e4 LT |
349 | |
350 | if (!pmd_present(*pmd_k)) | |
351 | goto bad_area_nosemaphore; | |
352 | ||
353 | set_pmd(pmd, *pmd_k); | |
354 | ||
355 | /* Make sure the actual PTE exists as well to | |
356 | * catch kernel vmalloc-area accesses to non-mapped | |
357 | * addresses. If we don't do this, this will just | |
358 | * silently loop forever. | |
359 | */ | |
360 | ||
361 | pte_k = pte_offset_kernel(pmd_k, address); | |
362 | if (!pte_present(*pte_k)) | |
363 | goto no_context; | |
364 | ||
365 | return; | |
366 | } | |
367 | } | |
4f18cfbf MS |
368 | |
369 | /* Find fixup code. */ | |
370 | int | |
371 | find_fixup_code(struct pt_regs *regs) | |
372 | { | |
373 | const struct exception_table_entry *fixup; | |
a90993c6 JN |
374 | /* in case of delay slot fault (v32) */ |
375 | unsigned long ip = (instruction_pointer(regs) & ~0x1); | |
4f18cfbf | 376 | |
a90993c6 JN |
377 | fixup = search_exception_tables(ip); |
378 | if (fixup != 0) { | |
4f18cfbf MS |
379 | /* Adjust the instruction pointer in the stackframe. */ |
380 | instruction_pointer(regs) = fixup->fixup; | |
381 | arch_fixup(regs); | |
382 | return 1; | |
383 | } | |
384 | ||
385 | return 0; | |
386 | } |