Loading...
1/*
2 * Page fault handler for SH with an MMU.
3 *
4 * Copyright (C) 1999 Niibe Yutaka
5 * Copyright (C) 2003 - 2012 Paul Mundt
6 *
7 * Based on linux/arch/i386/mm/fault.c:
8 * Copyright (C) 1995 Linus Torvalds
9 *
10 * This file is subject to the terms and conditions of the GNU General Public
11 * License. See the file "COPYING" in the main directory of this archive
12 * for more details.
13 */
14#include <linux/kernel.h>
15#include <linux/mm.h>
16#include <linux/sched/signal.h>
17#include <linux/hardirq.h>
18#include <linux/kprobes.h>
19#include <linux/perf_event.h>
20#include <linux/kdebug.h>
21#include <linux/uaccess.h>
22#include <asm/io_trapped.h>
23#include <asm/mmu_context.h>
24#include <asm/tlbflush.h>
25#include <asm/traps.h>
26
27static void
28force_sig_info_fault(int si_signo, int si_code, unsigned long address)
29{
30 force_sig_fault(si_signo, si_code, (void __user *)address);
31}
32
33/*
34 * This is useful to dump out the page tables associated with
35 * 'addr' in mm 'mm'.
36 */
37static void show_pte(struct mm_struct *mm, unsigned long addr)
38{
39 pgd_t *pgd;
40
41 if (mm) {
42 pgd = mm->pgd;
43 } else {
44 pgd = get_TTB();
45
46 if (unlikely(!pgd))
47 pgd = swapper_pg_dir;
48 }
49
50 printk(KERN_ALERT "pgd = %p\n", pgd);
51 pgd += pgd_index(addr);
52 printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
53 (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd));
54
55 do {
56 pud_t *pud;
57 pmd_t *pmd;
58 pte_t *pte;
59
60 if (pgd_none(*pgd))
61 break;
62
63 if (pgd_bad(*pgd)) {
64 printk("(bad)");
65 break;
66 }
67
68 pud = pud_offset(pgd, addr);
69 if (PTRS_PER_PUD != 1)
70 printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
71 (u64)pud_val(*pud));
72
73 if (pud_none(*pud))
74 break;
75
76 if (pud_bad(*pud)) {
77 printk("(bad)");
78 break;
79 }
80
81 pmd = pmd_offset(pud, addr);
82 if (PTRS_PER_PMD != 1)
83 printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
84 (u64)pmd_val(*pmd));
85
86 if (pmd_none(*pmd))
87 break;
88
89 if (pmd_bad(*pmd)) {
90 printk("(bad)");
91 break;
92 }
93
94 /* We must not map this if we have highmem enabled */
95 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
96 break;
97
98 pte = pte_offset_kernel(pmd, addr);
99 printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
100 (u64)pte_val(*pte));
101 } while (0);
102
103 printk("\n");
104}
105
106static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
107{
108 unsigned index = pgd_index(address);
109 pgd_t *pgd_k;
110 pud_t *pud, *pud_k;
111 pmd_t *pmd, *pmd_k;
112
113 pgd += index;
114 pgd_k = init_mm.pgd + index;
115
116 if (!pgd_present(*pgd_k))
117 return NULL;
118
119 pud = pud_offset(pgd, address);
120 pud_k = pud_offset(pgd_k, address);
121 if (!pud_present(*pud_k))
122 return NULL;
123
124 if (!pud_present(*pud))
125 set_pud(pud, *pud_k);
126
127 pmd = pmd_offset(pud, address);
128 pmd_k = pmd_offset(pud_k, address);
129 if (!pmd_present(*pmd_k))
130 return NULL;
131
132 if (!pmd_present(*pmd))
133 set_pmd(pmd, *pmd_k);
134 else {
135 /*
136 * The page tables are fully synchronised so there must
137 * be another reason for the fault. Return NULL here to
138 * signal that we have not taken care of the fault.
139 */
140 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
141 return NULL;
142 }
143
144 return pmd_k;
145}
146
147#ifdef CONFIG_SH_STORE_QUEUES
148#define __FAULT_ADDR_LIMIT P3_ADDR_MAX
149#else
150#define __FAULT_ADDR_LIMIT VMALLOC_END
151#endif
152
153/*
154 * Handle a fault on the vmalloc or module mapping area
155 */
156static noinline int vmalloc_fault(unsigned long address)
157{
158 pgd_t *pgd_k;
159 pmd_t *pmd_k;
160 pte_t *pte_k;
161
162 /* Make sure we are in vmalloc/module/P3 area: */
163 if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
164 return -1;
165
166 /*
167 * Synchronize this task's top level page-table
168 * with the 'reference' page table.
169 *
170 * Do _not_ use "current" here. We might be inside
171 * an interrupt in the middle of a task switch..
172 */
173 pgd_k = get_TTB();
174 pmd_k = vmalloc_sync_one(pgd_k, address);
175 if (!pmd_k)
176 return -1;
177
178 pte_k = pte_offset_kernel(pmd_k, address);
179 if (!pte_present(*pte_k))
180 return -1;
181
182 return 0;
183}
184
185static void
186show_fault_oops(struct pt_regs *regs, unsigned long address)
187{
188 if (!oops_may_print())
189 return;
190
191 printk(KERN_ALERT "BUG: unable to handle kernel ");
192 if (address < PAGE_SIZE)
193 printk(KERN_CONT "NULL pointer dereference");
194 else
195 printk(KERN_CONT "paging request");
196
197 printk(KERN_CONT " at %08lx\n", address);
198 printk(KERN_ALERT "PC:");
199 printk_address(regs->pc, 1);
200
201 show_pte(NULL, address);
202}
203
204static noinline void
205no_context(struct pt_regs *regs, unsigned long error_code,
206 unsigned long address)
207{
208 /* Are we prepared to handle this kernel fault? */
209 if (fixup_exception(regs))
210 return;
211
212 if (handle_trapped_io(regs, address))
213 return;
214
215 /*
216 * Oops. The kernel tried to access some bad page. We'll have to
217 * terminate things with extreme prejudice.
218 */
219 bust_spinlocks(1);
220
221 show_fault_oops(regs, address);
222
223 die("Oops", regs, error_code);
224 bust_spinlocks(0);
225 do_exit(SIGKILL);
226}
227
228static void
229__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
230 unsigned long address, int si_code)
231{
232 /* User mode accesses just cause a SIGSEGV */
233 if (user_mode(regs)) {
234 /*
235 * It's possible to have interrupts off here:
236 */
237 local_irq_enable();
238
239 force_sig_info_fault(SIGSEGV, si_code, address);
240
241 return;
242 }
243
244 no_context(regs, error_code, address);
245}
246
247static noinline void
248bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
249 unsigned long address)
250{
251 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
252}
253
254static void
255__bad_area(struct pt_regs *regs, unsigned long error_code,
256 unsigned long address, int si_code)
257{
258 struct mm_struct *mm = current->mm;
259
260 /*
261 * Something tried to access memory that isn't in our memory map..
262 * Fix it, but check if it's kernel or user first..
263 */
264 up_read(&mm->mmap_sem);
265
266 __bad_area_nosemaphore(regs, error_code, address, si_code);
267}
268
269static noinline void
270bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
271{
272 __bad_area(regs, error_code, address, SEGV_MAPERR);
273}
274
275static noinline void
276bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
277 unsigned long address)
278{
279 __bad_area(regs, error_code, address, SEGV_ACCERR);
280}
281
282static void
283do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
284{
285 struct task_struct *tsk = current;
286 struct mm_struct *mm = tsk->mm;
287
288 up_read(&mm->mmap_sem);
289
290 /* Kernel mode? Handle exceptions or die: */
291 if (!user_mode(regs))
292 no_context(regs, error_code, address);
293
294 force_sig_info_fault(SIGBUS, BUS_ADRERR, address);
295}
296
297static noinline int
298mm_fault_error(struct pt_regs *regs, unsigned long error_code,
299 unsigned long address, vm_fault_t fault)
300{
301 /*
302 * Pagefault was interrupted by SIGKILL. We have no reason to
303 * continue pagefault.
304 */
305 if (fatal_signal_pending(current)) {
306 if (!(fault & VM_FAULT_RETRY))
307 up_read(¤t->mm->mmap_sem);
308 if (!user_mode(regs))
309 no_context(regs, error_code, address);
310 return 1;
311 }
312
313 if (!(fault & VM_FAULT_ERROR))
314 return 0;
315
316 if (fault & VM_FAULT_OOM) {
317 /* Kernel mode? Handle exceptions or die: */
318 if (!user_mode(regs)) {
319 up_read(¤t->mm->mmap_sem);
320 no_context(regs, error_code, address);
321 return 1;
322 }
323 up_read(¤t->mm->mmap_sem);
324
325 /*
326 * We ran out of memory, call the OOM killer, and return the
327 * userspace (which will retry the fault, or kill us if we got
328 * oom-killed):
329 */
330 pagefault_out_of_memory();
331 } else {
332 if (fault & VM_FAULT_SIGBUS)
333 do_sigbus(regs, error_code, address);
334 else if (fault & VM_FAULT_SIGSEGV)
335 bad_area(regs, error_code, address);
336 else
337 BUG();
338 }
339
340 return 1;
341}
342
343static inline int access_error(int error_code, struct vm_area_struct *vma)
344{
345 if (error_code & FAULT_CODE_WRITE) {
346 /* write, present and write, not present: */
347 if (unlikely(!(vma->vm_flags & VM_WRITE)))
348 return 1;
349 return 0;
350 }
351
352 /* ITLB miss on NX page */
353 if (unlikely((error_code & FAULT_CODE_ITLB) &&
354 !(vma->vm_flags & VM_EXEC)))
355 return 1;
356
357 /* read, not present: */
358 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
359 return 1;
360
361 return 0;
362}
363
364static int fault_in_kernel_space(unsigned long address)
365{
366 return address >= TASK_SIZE;
367}
368
369/*
370 * This routine handles page faults. It determines the address,
371 * and the problem, and then passes it off to one of the appropriate
372 * routines.
373 */
374asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
375 unsigned long error_code,
376 unsigned long address)
377{
378 unsigned long vec;
379 struct task_struct *tsk;
380 struct mm_struct *mm;
381 struct vm_area_struct * vma;
382 vm_fault_t fault;
383 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
384
385 tsk = current;
386 mm = tsk->mm;
387 vec = lookup_exception_vector();
388
389 /*
390 * We fault-in kernel-space virtual memory on-demand. The
391 * 'reference' page table is init_mm.pgd.
392 *
393 * NOTE! We MUST NOT take any locks for this case. We may
394 * be in an interrupt or a critical region, and should
395 * only copy the information from the master page table,
396 * nothing more.
397 */
398 if (unlikely(fault_in_kernel_space(address))) {
399 if (vmalloc_fault(address) >= 0)
400 return;
401 if (kprobe_page_fault(regs, vec))
402 return;
403
404 bad_area_nosemaphore(regs, error_code, address);
405 return;
406 }
407
408 if (unlikely(kprobe_page_fault(regs, vec)))
409 return;
410
411 /* Only enable interrupts if they were on before the fault */
412 if ((regs->sr & SR_IMASK) != SR_IMASK)
413 local_irq_enable();
414
415 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
416
417 /*
418 * If we're in an interrupt, have no user context or are running
419 * with pagefaults disabled then we must not take the fault:
420 */
421 if (unlikely(faulthandler_disabled() || !mm)) {
422 bad_area_nosemaphore(regs, error_code, address);
423 return;
424 }
425
426retry:
427 down_read(&mm->mmap_sem);
428
429 vma = find_vma(mm, address);
430 if (unlikely(!vma)) {
431 bad_area(regs, error_code, address);
432 return;
433 }
434 if (likely(vma->vm_start <= address))
435 goto good_area;
436 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
437 bad_area(regs, error_code, address);
438 return;
439 }
440 if (unlikely(expand_stack(vma, address))) {
441 bad_area(regs, error_code, address);
442 return;
443 }
444
445 /*
446 * Ok, we have a good vm_area for this memory access, so
447 * we can handle it..
448 */
449good_area:
450 if (unlikely(access_error(error_code, vma))) {
451 bad_area_access_error(regs, error_code, address);
452 return;
453 }
454
455 set_thread_fault_code(error_code);
456
457 if (user_mode(regs))
458 flags |= FAULT_FLAG_USER;
459 if (error_code & FAULT_CODE_WRITE)
460 flags |= FAULT_FLAG_WRITE;
461
462 /*
463 * If for any reason at all we couldn't handle the fault,
464 * make sure we exit gracefully rather than endlessly redo
465 * the fault.
466 */
467 fault = handle_mm_fault(vma, address, flags);
468
469 if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
470 if (mm_fault_error(regs, error_code, address, fault))
471 return;
472
473 if (flags & FAULT_FLAG_ALLOW_RETRY) {
474 if (fault & VM_FAULT_MAJOR) {
475 tsk->maj_flt++;
476 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
477 regs, address);
478 } else {
479 tsk->min_flt++;
480 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
481 regs, address);
482 }
483 if (fault & VM_FAULT_RETRY) {
484 flags &= ~FAULT_FLAG_ALLOW_RETRY;
485 flags |= FAULT_FLAG_TRIED;
486
487 /*
488 * No need to up_read(&mm->mmap_sem) as we would
489 * have already released it in __lock_page_or_retry
490 * in mm/filemap.c.
491 */
492 goto retry;
493 }
494 }
495
496 up_read(&mm->mmap_sem);
497}
1/*
2 * Page fault handler for SH with an MMU.
3 *
4 * Copyright (C) 1999 Niibe Yutaka
5 * Copyright (C) 2003 - 2012 Paul Mundt
6 *
7 * Based on linux/arch/i386/mm/fault.c:
8 * Copyright (C) 1995 Linus Torvalds
9 *
10 * This file is subject to the terms and conditions of the GNU General Public
11 * License. See the file "COPYING" in the main directory of this archive
12 * for more details.
13 */
14#include <linux/kernel.h>
15#include <linux/mm.h>
16#include <linux/sched/signal.h>
17#include <linux/hardirq.h>
18#include <linux/kprobes.h>
19#include <linux/perf_event.h>
20#include <linux/kdebug.h>
21#include <linux/uaccess.h>
22#include <asm/io_trapped.h>
23#include <asm/mmu_context.h>
24#include <asm/tlbflush.h>
25#include <asm/traps.h>
26
27static void
28force_sig_info_fault(int si_signo, int si_code, unsigned long address)
29{
30 force_sig_fault(si_signo, si_code, (void __user *)address);
31}
32
33/*
34 * This is useful to dump out the page tables associated with
35 * 'addr' in mm 'mm'.
36 */
37static void show_pte(struct mm_struct *mm, unsigned long addr)
38{
39 pgd_t *pgd;
40
41 if (mm) {
42 pgd = mm->pgd;
43 } else {
44 pgd = get_TTB();
45
46 if (unlikely(!pgd))
47 pgd = swapper_pg_dir;
48 }
49
50 pr_alert("pgd = %p\n", pgd);
51 pgd += pgd_index(addr);
52 pr_alert("[%08lx] *pgd=%0*llx", addr, (u32)(sizeof(*pgd) * 2),
53 (u64)pgd_val(*pgd));
54
55 do {
56 p4d_t *p4d;
57 pud_t *pud;
58 pmd_t *pmd;
59 pte_t *pte;
60
61 if (pgd_none(*pgd))
62 break;
63
64 if (pgd_bad(*pgd)) {
65 pr_cont("(bad)");
66 break;
67 }
68
69 p4d = p4d_offset(pgd, addr);
70 if (PTRS_PER_P4D != 1)
71 pr_cont(", *p4d=%0*Lx", (u32)(sizeof(*p4d) * 2),
72 (u64)p4d_val(*p4d));
73
74 if (p4d_none(*p4d))
75 break;
76
77 if (p4d_bad(*p4d)) {
78 pr_cont("(bad)");
79 break;
80 }
81
82 pud = pud_offset(p4d, addr);
83 if (PTRS_PER_PUD != 1)
84 pr_cont(", *pud=%0*llx", (u32)(sizeof(*pud) * 2),
85 (u64)pud_val(*pud));
86
87 if (pud_none(*pud))
88 break;
89
90 if (pud_bad(*pud)) {
91 pr_cont("(bad)");
92 break;
93 }
94
95 pmd = pmd_offset(pud, addr);
96 if (PTRS_PER_PMD != 1)
97 pr_cont(", *pmd=%0*llx", (u32)(sizeof(*pmd) * 2),
98 (u64)pmd_val(*pmd));
99
100 if (pmd_none(*pmd))
101 break;
102
103 if (pmd_bad(*pmd)) {
104 pr_cont("(bad)");
105 break;
106 }
107
108 /* We must not map this if we have highmem enabled */
109 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
110 break;
111
112 pte = pte_offset_kernel(pmd, addr);
113 pr_cont(", *pte=%0*llx", (u32)(sizeof(*pte) * 2),
114 (u64)pte_val(*pte));
115 } while (0);
116
117 pr_cont("\n");
118}
119
120static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
121{
122 unsigned index = pgd_index(address);
123 pgd_t *pgd_k;
124 p4d_t *p4d, *p4d_k;
125 pud_t *pud, *pud_k;
126 pmd_t *pmd, *pmd_k;
127
128 pgd += index;
129 pgd_k = init_mm.pgd + index;
130
131 if (!pgd_present(*pgd_k))
132 return NULL;
133
134 p4d = p4d_offset(pgd, address);
135 p4d_k = p4d_offset(pgd_k, address);
136 if (!p4d_present(*p4d_k))
137 return NULL;
138
139 pud = pud_offset(p4d, address);
140 pud_k = pud_offset(p4d_k, address);
141 if (!pud_present(*pud_k))
142 return NULL;
143
144 if (!pud_present(*pud))
145 set_pud(pud, *pud_k);
146
147 pmd = pmd_offset(pud, address);
148 pmd_k = pmd_offset(pud_k, address);
149 if (!pmd_present(*pmd_k))
150 return NULL;
151
152 if (!pmd_present(*pmd))
153 set_pmd(pmd, *pmd_k);
154 else {
155 /*
156 * The page tables are fully synchronised so there must
157 * be another reason for the fault. Return NULL here to
158 * signal that we have not taken care of the fault.
159 */
160 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
161 return NULL;
162 }
163
164 return pmd_k;
165}
166
167#ifdef CONFIG_SH_STORE_QUEUES
168#define __FAULT_ADDR_LIMIT P3_ADDR_MAX
169#else
170#define __FAULT_ADDR_LIMIT VMALLOC_END
171#endif
172
173/*
174 * Handle a fault on the vmalloc or module mapping area
175 */
176static noinline int vmalloc_fault(unsigned long address)
177{
178 pgd_t *pgd_k;
179 pmd_t *pmd_k;
180 pte_t *pte_k;
181
182 /* Make sure we are in vmalloc/module/P3 area: */
183 if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
184 return -1;
185
186 /*
187 * Synchronize this task's top level page-table
188 * with the 'reference' page table.
189 *
190 * Do _not_ use "current" here. We might be inside
191 * an interrupt in the middle of a task switch..
192 */
193 pgd_k = get_TTB();
194 pmd_k = vmalloc_sync_one(pgd_k, address);
195 if (!pmd_k)
196 return -1;
197
198 pte_k = pte_offset_kernel(pmd_k, address);
199 if (!pte_present(*pte_k))
200 return -1;
201
202 return 0;
203}
204
205static void
206show_fault_oops(struct pt_regs *regs, unsigned long address)
207{
208 if (!oops_may_print())
209 return;
210
211 pr_alert("BUG: unable to handle kernel %s at %08lx\n",
212 address < PAGE_SIZE ? "NULL pointer dereference"
213 : "paging request",
214 address);
215 pr_alert("PC:");
216 printk_address(regs->pc, 1);
217
218 show_pte(NULL, address);
219}
220
221static noinline void
222no_context(struct pt_regs *regs, unsigned long error_code,
223 unsigned long address)
224{
225 /* Are we prepared to handle this kernel fault? */
226 if (fixup_exception(regs))
227 return;
228
229 if (handle_trapped_io(regs, address))
230 return;
231
232 /*
233 * Oops. The kernel tried to access some bad page. We'll have to
234 * terminate things with extreme prejudice.
235 */
236 bust_spinlocks(1);
237
238 show_fault_oops(regs, address);
239
240 die("Oops", regs, error_code);
241}
242
243static void
244__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
245 unsigned long address, int si_code)
246{
247 /* User mode accesses just cause a SIGSEGV */
248 if (user_mode(regs)) {
249 /*
250 * It's possible to have interrupts off here:
251 */
252 local_irq_enable();
253
254 force_sig_info_fault(SIGSEGV, si_code, address);
255
256 return;
257 }
258
259 no_context(regs, error_code, address);
260}
261
262static noinline void
263bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
264 unsigned long address)
265{
266 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
267}
268
269static void
270__bad_area(struct pt_regs *regs, unsigned long error_code,
271 unsigned long address, int si_code)
272{
273 struct mm_struct *mm = current->mm;
274
275 /*
276 * Something tried to access memory that isn't in our memory map..
277 * Fix it, but check if it's kernel or user first..
278 */
279 mmap_read_unlock(mm);
280
281 __bad_area_nosemaphore(regs, error_code, address, si_code);
282}
283
284static noinline void
285bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
286{
287 __bad_area(regs, error_code, address, SEGV_MAPERR);
288}
289
290static noinline void
291bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
292 unsigned long address)
293{
294 __bad_area(regs, error_code, address, SEGV_ACCERR);
295}
296
297static void
298do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
299{
300 struct task_struct *tsk = current;
301 struct mm_struct *mm = tsk->mm;
302
303 mmap_read_unlock(mm);
304
305 /* Kernel mode? Handle exceptions or die: */
306 if (!user_mode(regs))
307 no_context(regs, error_code, address);
308
309 force_sig_info_fault(SIGBUS, BUS_ADRERR, address);
310}
311
312static noinline int
313mm_fault_error(struct pt_regs *regs, unsigned long error_code,
314 unsigned long address, vm_fault_t fault)
315{
316 /*
317 * Pagefault was interrupted by SIGKILL. We have no reason to
318 * continue pagefault.
319 */
320 if (fault_signal_pending(fault, regs)) {
321 if (!user_mode(regs))
322 no_context(regs, error_code, address);
323 return 1;
324 }
325
326 /* Release mmap_lock first if necessary */
327 if (!(fault & VM_FAULT_RETRY))
328 mmap_read_unlock(current->mm);
329
330 if (!(fault & VM_FAULT_ERROR))
331 return 0;
332
333 if (fault & VM_FAULT_OOM) {
334 /* Kernel mode? Handle exceptions or die: */
335 if (!user_mode(regs)) {
336 no_context(regs, error_code, address);
337 return 1;
338 }
339
340 /*
341 * We ran out of memory, call the OOM killer, and return the
342 * userspace (which will retry the fault, or kill us if we got
343 * oom-killed):
344 */
345 pagefault_out_of_memory();
346 } else {
347 if (fault & VM_FAULT_SIGBUS)
348 do_sigbus(regs, error_code, address);
349 else if (fault & VM_FAULT_SIGSEGV)
350 bad_area(regs, error_code, address);
351 else
352 BUG();
353 }
354
355 return 1;
356}
357
358static inline int access_error(int error_code, struct vm_area_struct *vma)
359{
360 if (error_code & FAULT_CODE_WRITE) {
361 /* write, present and write, not present: */
362 if (unlikely(!(vma->vm_flags & VM_WRITE)))
363 return 1;
364 return 0;
365 }
366
367 /* ITLB miss on NX page */
368 if (unlikely((error_code & FAULT_CODE_ITLB) &&
369 !(vma->vm_flags & VM_EXEC)))
370 return 1;
371
372 /* read, not present: */
373 if (unlikely(!vma_is_accessible(vma)))
374 return 1;
375
376 return 0;
377}
378
379static int fault_in_kernel_space(unsigned long address)
380{
381 return address >= TASK_SIZE;
382}
383
384/*
385 * This routine handles page faults. It determines the address,
386 * and the problem, and then passes it off to one of the appropriate
387 * routines.
388 */
389asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
390 unsigned long error_code,
391 unsigned long address)
392{
393 unsigned long vec;
394 struct task_struct *tsk;
395 struct mm_struct *mm;
396 struct vm_area_struct * vma;
397 vm_fault_t fault;
398 unsigned int flags = FAULT_FLAG_DEFAULT;
399
400 tsk = current;
401 mm = tsk->mm;
402 vec = lookup_exception_vector();
403
404 /*
405 * We fault-in kernel-space virtual memory on-demand. The
406 * 'reference' page table is init_mm.pgd.
407 *
408 * NOTE! We MUST NOT take any locks for this case. We may
409 * be in an interrupt or a critical region, and should
410 * only copy the information from the master page table,
411 * nothing more.
412 */
413 if (unlikely(fault_in_kernel_space(address))) {
414 if (vmalloc_fault(address) >= 0)
415 return;
416 if (kprobe_page_fault(regs, vec))
417 return;
418
419 bad_area_nosemaphore(regs, error_code, address);
420 return;
421 }
422
423 if (unlikely(kprobe_page_fault(regs, vec)))
424 return;
425
426 /* Only enable interrupts if they were on before the fault */
427 if ((regs->sr & SR_IMASK) != SR_IMASK)
428 local_irq_enable();
429
430 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
431
432 /*
433 * If we're in an interrupt, have no user context or are running
434 * with pagefaults disabled then we must not take the fault:
435 */
436 if (unlikely(faulthandler_disabled() || !mm)) {
437 bad_area_nosemaphore(regs, error_code, address);
438 return;
439 }
440
441retry:
442 vma = lock_mm_and_find_vma(mm, address, regs);
443 if (unlikely(!vma)) {
444 bad_area_nosemaphore(regs, error_code, address);
445 return;
446 }
447
448 /*
449 * Ok, we have a good vm_area for this memory access, so
450 * we can handle it..
451 */
452 if (unlikely(access_error(error_code, vma))) {
453 bad_area_access_error(regs, error_code, address);
454 return;
455 }
456
457 set_thread_fault_code(error_code);
458
459 if (user_mode(regs))
460 flags |= FAULT_FLAG_USER;
461 if (error_code & FAULT_CODE_WRITE)
462 flags |= FAULT_FLAG_WRITE;
463
464 /*
465 * If for any reason at all we couldn't handle the fault,
466 * make sure we exit gracefully rather than endlessly redo
467 * the fault.
468 */
469 fault = handle_mm_fault(vma, address, flags, regs);
470
471 if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
472 if (mm_fault_error(regs, error_code, address, fault))
473 return;
474
475 /* The fault is fully completed (including releasing mmap lock) */
476 if (fault & VM_FAULT_COMPLETED)
477 return;
478
479 if (fault & VM_FAULT_RETRY) {
480 flags |= FAULT_FLAG_TRIED;
481
482 /*
483 * No need to mmap_read_unlock(mm) as we would
484 * have already released it in __lock_page_or_retry
485 * in mm/filemap.c.
486 */
487 goto retry;
488 }
489
490 mmap_read_unlock(mm);
491}