1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (C) 2005-2017 Andes Technology Corporation
3
4 #include <linux/extable.h>
5 #include <linux/module.h>
6 #include <linux/signal.h>
7 #include <linux/ptrace.h>
8 #include <linux/mm.h>
9 #include <linux/init.h>
10 #include <linux/hardirq.h>
11 #include <linux/uaccess.h>
12 #include <linux/perf_event.h>
13
14 #include <asm/pgtable.h>
15 #include <asm/tlbflush.h>
16
17 extern void die(const char *str, struct pt_regs *regs, long err);
18
19 /*
20 * This is useful to dump out the page tables associated with
21 * 'addr' in mm 'mm'.
22 */
23 void show_pte(struct mm_struct *mm, unsigned long addr)
24 {
25 pgd_t *pgd;
26 if (!mm)
27 mm = &init_mm;
28
29 pr_alert("pgd = %p\n", mm->pgd);
30 pgd = pgd_offset(mm, addr);
31 pr_alert("[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
32
33 do {
34 p4d_t *p4d;
35 pud_t *pud;
36 pmd_t *pmd;
37
38 if (pgd_none(*pgd))
39 break;
40
41 if (pgd_bad(*pgd)) {
42 pr_alert("(bad)");
43 break;
44 }
45
46 p4d = p4d_offset(pgd, addr);
47 pud = pud_offset(p4d, addr);
48 pmd = pmd_offset(pud, addr);
49 #if PTRS_PER_PMD != 1
50 pr_alert(", *pmd=%08lx", pmd_val(*pmd));
51 #endif
52
53 if (pmd_none(*pmd))
54 break;
55
56 if (pmd_bad(*pmd)) {
57 pr_alert("(bad)");
58 break;
59 }
60
61 if (IS_ENABLED(CONFIG_HIGHMEM))
62 {
63 pte_t *pte;
64 /* We must not map this if we have highmem enabled */
65 pte = pte_offset_map(pmd, addr);
66 pr_alert(", *pte=%08lx", pte_val(*pte));
67 pte_unmap(pte);
68 }
69 } while (0);
70
71 pr_alert("\n");
72 }
73
74 void do_page_fault(unsigned long entry, unsigned long addr,
75 unsigned int error_code, struct pt_regs *regs)
76 {
77 struct task_struct *tsk;
78 struct mm_struct *mm;
79 struct vm_area_struct *vma;
80 int si_code;
81 vm_fault_t fault;
82 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
83 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
84
85 error_code = error_code & (ITYPE_mskINST | ITYPE_mskETYPE);
86 tsk = current;
87 mm = tsk->mm;
88 si_code = SEGV_MAPERR;
89 /*
90 * We fault-in kernel-space virtual memory on-demand. The
91 * 'reference' page table is init_mm.pgd.
92 *
93 * NOTE! We MUST NOT take any locks for this case. We may
94 * be in an interrupt or a critical region, and should
95 * only copy the information from the master page table,
96 * nothing more.
97 */
98 if (addr >= TASK_SIZE) {
99 if (user_mode(regs))
100 goto bad_area_nosemaphore;
101
102 if (addr >= TASK_SIZE && addr < VMALLOC_END
103 && (entry == ENTRY_PTE_NOT_PRESENT))
104 goto vmalloc_fault;
105 else
106 goto no_context;
107 }
108
109 /* Send a signal to the task for handling the unalignment access. */
110 if (entry == ENTRY_GENERAL_EXCPETION
111 && error_code == ETYPE_ALIGNMENT_CHECK) {
112 if (user_mode(regs))
113 goto bad_area_nosemaphore;
114 else
115 goto no_context;
116 }
117
118 /*
119 * If we're in an interrupt or have no user
120 * context, we must not take the fault..
121 */
122 if (unlikely(faulthandler_disabled() || !mm))
123 goto no_context;
124
125 /*
126 * As per x86, we may deadlock here. However, since the kernel only
127 * validly references user space from well defined areas of the code,
128 * we can bug out early if this is from code which shouldn't.
129 */
130 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
131 if (!user_mode(regs) &&
132 !search_exception_tables(instruction_pointer(regs)))
133 goto no_context;
134 retry:
135 down_read(&mm->mmap_sem);
136 } else {
137 /*
138 * The above down_read_trylock() might have succeeded in which
139 * case, we'll have missed the might_sleep() from down_read().
140 */
141 might_sleep();
142 if (IS_ENABLED(CONFIG_DEBUG_VM)) {
143 if (!user_mode(regs) &&
144 !search_exception_tables(instruction_pointer(regs)))
145 goto no_context;
146 }
147 }
148
149 vma = find_vma(mm, addr);
150
151 if (unlikely(!vma))
152 goto bad_area;
153
154 if (vma->vm_start <= addr)
155 goto good_area;
156
157 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
158 goto bad_area;
159
160 if (unlikely(expand_stack(vma, addr)))
161 goto bad_area;
162
163 /*
164 * Ok, we have a good vm_area for this memory access, so
165 * we can handle it..
166 */
167
168 good_area:
169 si_code = SEGV_ACCERR;
170
171 /* first do some preliminary protection checks */
172 if (entry == ENTRY_PTE_NOT_PRESENT) {
173 if (error_code & ITYPE_mskINST)
174 mask = VM_EXEC;
175 else {
176 mask = VM_READ | VM_WRITE;
177 }
178 } else if (entry == ENTRY_TLB_MISC) {
179 switch (error_code & ITYPE_mskETYPE) {
180 case RD_PROT:
181 mask = VM_READ;
182 break;
183 case WRT_PROT:
184 mask = VM_WRITE;
185 flags |= FAULT_FLAG_WRITE;
186 break;
187 case NOEXEC:
188 mask = VM_EXEC;
189 break;
190 case PAGE_MODIFY:
191 mask = VM_WRITE;
192 flags |= FAULT_FLAG_WRITE;
193 break;
194 case ACC_BIT:
195 BUG();
196 default:
197 break;
198 }
199
200 }
201 if (!(vma->vm_flags & mask))
202 goto bad_area;
203
204 /*
205 * If for any reason at all we couldn't handle the fault,
206 * make sure we exit gracefully rather than endlessly redo
207 * the fault.
208 */
209
210 fault = handle_mm_fault(vma, addr, flags);
211
212 /*
213 * If we need to retry but a fatal signal is pending, handle the
214 * signal first. We do not need to release the mmap_sem because it
215 * would already be released in __lock_page_or_retry in mm/filemap.c.
216 */
217 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
218 if (!user_mode(regs))
219 goto no_context;
220 return;
221 }
222
223 if (unlikely(fault & VM_FAULT_ERROR)) {
224 if (fault & VM_FAULT_OOM)
225 goto out_of_memory;
226 else if (fault & VM_FAULT_SIGBUS)
227 goto do_sigbus;
228 else
229 goto bad_area;
230 }
231
232 /*
233 * Major/minor page fault accounting is only done on the initial
234 * attempt. If we go through a retry, it is extremely likely that the
235 * page will be found in page cache at that point.
236 */
237 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
238 if (flags & FAULT_FLAG_ALLOW_RETRY) {
239 if (fault & VM_FAULT_MAJOR) {
240 tsk->maj_flt++;
241 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
242 1, regs, addr);
243 } else {
244 tsk->min_flt++;
245 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
246 1, regs, addr);
247 }
248 if (fault & VM_FAULT_RETRY) {
249 flags &= ~FAULT_FLAG_ALLOW_RETRY;
250 flags |= FAULT_FLAG_TRIED;
251
252 /* No need to up_read(&mm->mmap_sem) as we would
253 * have already released it in __lock_page_or_retry
254 * in mm/filemap.c.
255 */
256 goto retry;
257 }
258 }
259
260 up_read(&mm->mmap_sem);
261 return;
262
263 /*
264 * Something tried to access memory that isn't in our memory map..
265 * Fix it, but check if it's kernel or user first..
266 */
267 bad_area:
268 up_read(&mm->mmap_sem);
269
270 bad_area_nosemaphore:
271
272 /* User mode accesses just cause a SIGSEGV */
273
274 if (user_mode(regs)) {
275 tsk->thread.address = addr;
276 tsk->thread.error_code = error_code;
277 tsk->thread.trap_no = entry;
278 force_sig_fault(SIGSEGV, si_code, (void __user *)addr);
279 return;
280 }
281
282 no_context:
283
284 /* Are we prepared to handle this kernel fault?
285 *
286 * (The kernel has valid exception-points in the source
287 * when it acesses user-memory. When it fails in one
288 * of those points, we find it in a table and do a jump
289 * to some fixup code that loads an appropriate error
290 * code)
291 */
292
293 {
294 const struct exception_table_entry *entry;
295
296 if ((entry =
297 search_exception_tables(instruction_pointer(regs))) !=
298 NULL) {
299 /* Adjust the instruction pointer in the stackframe */
300 instruction_pointer(regs) = entry->fixup;
301 return;
302 }
303 }
304
305 /*
306 * Oops. The kernel tried to access some bad page. We'll have to
307 * terminate things with extreme prejudice.
308 */
309
310 bust_spinlocks(1);
311 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
312 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
313 "paging request", addr);
314
315 show_pte(mm, addr);
316 die("Oops", regs, error_code);
317 bust_spinlocks(0);
318 do_exit(SIGKILL);
319
320 return;
321
322 /*
323 * We ran out of memory, or some other thing happened to us that made
324 * us unable to handle the page fault gracefully.
325 */
326
327 out_of_memory:
328 up_read(&mm->mmap_sem);
329 if (!user_mode(regs))
330 goto no_context;
331 pagefault_out_of_memory();
332 return;
333
334 do_sigbus:
335 up_read(&mm->mmap_sem);
336
337 /* Kernel mode? Handle exceptions or die */
338 if (!user_mode(regs))
339 goto no_context;
340
341 /*
342 * Send a sigbus
343 */
344 tsk->thread.address = addr;
345 tsk->thread.error_code = error_code;
346 tsk->thread.trap_no = entry;
347 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr);
348
349 return;
350
351 vmalloc_fault:
352 {
353 /*
354 * Synchronize this task's top level page-table
355 * with the 'reference' page table.
356 *
357 * Use current_pgd instead of tsk->active_mm->pgd
358 * since the latter might be unavailable if this
359 * code is executed in a misfortunately run irq
360 * (like inside schedule() between switch_mm and
361 * switch_to...).
362 */
363
364 unsigned int index = pgd_index(addr);
365 pgd_t *pgd, *pgd_k;
366 p4d_t *p4d, *p4d_k;
367 pud_t *pud, *pud_k;
368 pmd_t *pmd, *pmd_k;
369 pte_t *pte_k;
370
371 pgd = (pgd_t *) __va(__nds32__mfsr(NDS32_SR_L1_PPTB)) + index;
372 pgd_k = init_mm.pgd + index;
373
374 if (!pgd_present(*pgd_k))
375 goto no_context;
376
377 p4d = p4d_offset(pgd, addr);
378 p4d_k = p4d_offset(pgd_k, addr);
379 if (!p4d_present(*p4d_k))
380 goto no_context;
381
382 pud = pud_offset(p4d, addr);
383 pud_k = pud_offset(p4d_k, addr);
384 if (!pud_present(*pud_k))
385 goto no_context;
386
387 pmd = pmd_offset(pud, addr);
388 pmd_k = pmd_offset(pud_k, addr);
389 if (!pmd_present(*pmd_k))
390 goto no_context;
391
392 if (!pmd_present(*pmd))
393 set_pmd(pmd, *pmd_k);
394 else
395 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
396
397 /*
398 * Since the vmalloc area is global, we don't
399 * need to copy individual PTE's, it is enough to
400 * copy the pgd pointer into the pte page of the
401 * root task. If that is there, we'll find our pte if
402 * it exists.
403 */
404
405 /* Make sure the actual PTE exists as well to
406 * catch kernel vmalloc-area accesses to non-mapped
407 * addres. If we don't do this, this will just
408 * silently loop forever.
409 */
410
411 pte_k = pte_offset_kernel(pmd_k, addr);
412 if (!pte_present(*pte_k))
413 goto no_context;
414
415 return;
416 }
417 }
418
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.