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 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 mmap_read_lock(mm);
443
444 vma = find_vma(mm, address);
445 if (unlikely(!vma)) {
446 bad_area(regs, error_code, address);
447 return;
448 }
449 if (likely(vma->vm_start <= address))
450 goto good_area;
451 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
452 bad_area(regs, error_code, address);
453 return;
454 }
455 if (unlikely(expand_stack(vma, address))) {
456 bad_area(regs, error_code, address);
457 return;
458 }
459
460 /*
461 * Ok, we have a good vm_area for this memory access, so
462 * we can handle it..
463 */
464good_area:
465 if (unlikely(access_error(error_code, vma))) {
466 bad_area_access_error(regs, error_code, address);
467 return;
468 }
469
470 set_thread_fault_code(error_code);
471
472 if (user_mode(regs))
473 flags |= FAULT_FLAG_USER;
474 if (error_code & FAULT_CODE_WRITE)
475 flags |= FAULT_FLAG_WRITE;
476
477 /*
478 * If for any reason at all we couldn't handle the fault,
479 * make sure we exit gracefully rather than endlessly redo
480 * the fault.
481 */
482 fault = handle_mm_fault(vma, address, flags, regs);
483
484 if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
485 if (mm_fault_error(regs, error_code, address, fault))
486 return;
487
488 /* The fault is fully completed (including releasing mmap lock) */
489 if (fault & VM_FAULT_COMPLETED)
490 return;
491
492 if (fault & VM_FAULT_RETRY) {
493 flags |= FAULT_FLAG_TRIED;
494
495 /*
496 * No need to mmap_read_unlock(mm) as we would
497 * have already released it in __lock_page_or_retry
498 * in mm/filemap.c.
499 */
500 goto retry;
501 }
502
503 mmap_read_unlock(mm);
504}
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/hardirq.h>
17#include <linux/kprobes.h>
18#include <linux/perf_event.h>
19#include <linux/kdebug.h>
20#include <asm/io_trapped.h>
21#include <asm/mmu_context.h>
22#include <asm/tlbflush.h>
23#include <asm/traps.h>
24
25static inline int notify_page_fault(struct pt_regs *regs, int trap)
26{
27 int ret = 0;
28
29 if (kprobes_built_in() && !user_mode(regs)) {
30 preempt_disable();
31 if (kprobe_running() && kprobe_fault_handler(regs, trap))
32 ret = 1;
33 preempt_enable();
34 }
35
36 return ret;
37}
38
39static void
40force_sig_info_fault(int si_signo, int si_code, unsigned long address,
41 struct task_struct *tsk)
42{
43 siginfo_t info;
44
45 info.si_signo = si_signo;
46 info.si_errno = 0;
47 info.si_code = si_code;
48 info.si_addr = (void __user *)address;
49
50 force_sig_info(si_signo, &info, tsk);
51}
52
53/*
54 * This is useful to dump out the page tables associated with
55 * 'addr' in mm 'mm'.
56 */
57static void show_pte(struct mm_struct *mm, unsigned long addr)
58{
59 pgd_t *pgd;
60
61 if (mm) {
62 pgd = mm->pgd;
63 } else {
64 pgd = get_TTB();
65
66 if (unlikely(!pgd))
67 pgd = swapper_pg_dir;
68 }
69
70 printk(KERN_ALERT "pgd = %p\n", pgd);
71 pgd += pgd_index(addr);
72 printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
73 (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd));
74
75 do {
76 pud_t *pud;
77 pmd_t *pmd;
78 pte_t *pte;
79
80 if (pgd_none(*pgd))
81 break;
82
83 if (pgd_bad(*pgd)) {
84 printk("(bad)");
85 break;
86 }
87
88 pud = pud_offset(pgd, addr);
89 if (PTRS_PER_PUD != 1)
90 printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
91 (u64)pud_val(*pud));
92
93 if (pud_none(*pud))
94 break;
95
96 if (pud_bad(*pud)) {
97 printk("(bad)");
98 break;
99 }
100
101 pmd = pmd_offset(pud, addr);
102 if (PTRS_PER_PMD != 1)
103 printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
104 (u64)pmd_val(*pmd));
105
106 if (pmd_none(*pmd))
107 break;
108
109 if (pmd_bad(*pmd)) {
110 printk("(bad)");
111 break;
112 }
113
114 /* We must not map this if we have highmem enabled */
115 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
116 break;
117
118 pte = pte_offset_kernel(pmd, addr);
119 printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
120 (u64)pte_val(*pte));
121 } while (0);
122
123 printk("\n");
124}
125
126static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
127{
128 unsigned index = pgd_index(address);
129 pgd_t *pgd_k;
130 pud_t *pud, *pud_k;
131 pmd_t *pmd, *pmd_k;
132
133 pgd += index;
134 pgd_k = init_mm.pgd + index;
135
136 if (!pgd_present(*pgd_k))
137 return NULL;
138
139 pud = pud_offset(pgd, address);
140 pud_k = pud_offset(pgd_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 printk(KERN_ALERT "BUG: unable to handle kernel ");
212 if (address < PAGE_SIZE)
213 printk(KERN_CONT "NULL pointer dereference");
214 else
215 printk(KERN_CONT "paging request");
216
217 printk(KERN_CONT " at %08lx\n", address);
218 printk(KERN_ALERT "PC:");
219 printk_address(regs->pc, 1);
220
221 show_pte(NULL, address);
222}
223
224static noinline void
225no_context(struct pt_regs *regs, unsigned long error_code,
226 unsigned long address)
227{
228 /* Are we prepared to handle this kernel fault? */
229 if (fixup_exception(regs))
230 return;
231
232 if (handle_trapped_io(regs, address))
233 return;
234
235 /*
236 * Oops. The kernel tried to access some bad page. We'll have to
237 * terminate things with extreme prejudice.
238 */
239 bust_spinlocks(1);
240
241 show_fault_oops(regs, address);
242
243 die("Oops", regs, error_code);
244 bust_spinlocks(0);
245 do_exit(SIGKILL);
246}
247
248static void
249__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
250 unsigned long address, int si_code)
251{
252 struct task_struct *tsk = current;
253
254 /* User mode accesses just cause a SIGSEGV */
255 if (user_mode(regs)) {
256 /*
257 * It's possible to have interrupts off here:
258 */
259 local_irq_enable();
260
261 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
262
263 return;
264 }
265
266 no_context(regs, error_code, address);
267}
268
269static noinline void
270bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
271 unsigned long address)
272{
273 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
274}
275
276static void
277__bad_area(struct pt_regs *regs, unsigned long error_code,
278 unsigned long address, int si_code)
279{
280 struct mm_struct *mm = current->mm;
281
282 /*
283 * Something tried to access memory that isn't in our memory map..
284 * Fix it, but check if it's kernel or user first..
285 */
286 up_read(&mm->mmap_sem);
287
288 __bad_area_nosemaphore(regs, error_code, address, si_code);
289}
290
291static noinline void
292bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
293{
294 __bad_area(regs, error_code, address, SEGV_MAPERR);
295}
296
297static noinline void
298bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
299 unsigned long address)
300{
301 __bad_area(regs, error_code, address, SEGV_ACCERR);
302}
303
304static void
305do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
306{
307 struct task_struct *tsk = current;
308 struct mm_struct *mm = tsk->mm;
309
310 up_read(&mm->mmap_sem);
311
312 /* Kernel mode? Handle exceptions or die: */
313 if (!user_mode(regs))
314 no_context(regs, error_code, address);
315
316 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
317}
318
319static noinline int
320mm_fault_error(struct pt_regs *regs, unsigned long error_code,
321 unsigned long address, unsigned int fault)
322{
323 /*
324 * Pagefault was interrupted by SIGKILL. We have no reason to
325 * continue pagefault.
326 */
327 if (fatal_signal_pending(current)) {
328 if (!(fault & VM_FAULT_RETRY))
329 up_read(¤t->mm->mmap_sem);
330 if (!user_mode(regs))
331 no_context(regs, error_code, address);
332 return 1;
333 }
334
335 if (!(fault & VM_FAULT_ERROR))
336 return 0;
337
338 if (fault & VM_FAULT_OOM) {
339 /* Kernel mode? Handle exceptions or die: */
340 if (!user_mode(regs)) {
341 up_read(¤t->mm->mmap_sem);
342 no_context(regs, error_code, address);
343 return 1;
344 }
345 up_read(¤t->mm->mmap_sem);
346
347 /*
348 * We ran out of memory, call the OOM killer, and return the
349 * userspace (which will retry the fault, or kill us if we got
350 * oom-killed):
351 */
352 pagefault_out_of_memory();
353 } else {
354 if (fault & VM_FAULT_SIGBUS)
355 do_sigbus(regs, error_code, address);
356 else
357 BUG();
358 }
359
360 return 1;
361}
362
363static inline int access_error(int error_code, struct vm_area_struct *vma)
364{
365 if (error_code & FAULT_CODE_WRITE) {
366 /* write, present and write, not present: */
367 if (unlikely(!(vma->vm_flags & VM_WRITE)))
368 return 1;
369 return 0;
370 }
371
372 /* ITLB miss on NX page */
373 if (unlikely((error_code & FAULT_CODE_ITLB) &&
374 !(vma->vm_flags & VM_EXEC)))
375 return 1;
376
377 /* read, not present: */
378 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
379 return 1;
380
381 return 0;
382}
383
384static int fault_in_kernel_space(unsigned long address)
385{
386 return address >= TASK_SIZE;
387}
388
389/*
390 * This routine handles page faults. It determines the address,
391 * and the problem, and then passes it off to one of the appropriate
392 * routines.
393 */
394asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
395 unsigned long error_code,
396 unsigned long address)
397{
398 unsigned long vec;
399 struct task_struct *tsk;
400 struct mm_struct *mm;
401 struct vm_area_struct * vma;
402 int fault;
403 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
404
405 tsk = current;
406 mm = tsk->mm;
407 vec = lookup_exception_vector();
408
409 /*
410 * We fault-in kernel-space virtual memory on-demand. The
411 * 'reference' page table is init_mm.pgd.
412 *
413 * NOTE! We MUST NOT take any locks for this case. We may
414 * be in an interrupt or a critical region, and should
415 * only copy the information from the master page table,
416 * nothing more.
417 */
418 if (unlikely(fault_in_kernel_space(address))) {
419 if (vmalloc_fault(address) >= 0)
420 return;
421 if (notify_page_fault(regs, vec))
422 return;
423
424 bad_area_nosemaphore(regs, error_code, address);
425 return;
426 }
427
428 if (unlikely(notify_page_fault(regs, vec)))
429 return;
430
431 /* Only enable interrupts if they were on before the fault */
432 if ((regs->sr & SR_IMASK) != SR_IMASK)
433 local_irq_enable();
434
435 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
436
437 /*
438 * If we're in an interrupt, have no user context or are running
439 * in an atomic region then we must not take the fault:
440 */
441 if (unlikely(in_atomic() || !mm)) {
442 bad_area_nosemaphore(regs, error_code, address);
443 return;
444 }
445
446retry:
447 down_read(&mm->mmap_sem);
448
449 vma = find_vma(mm, address);
450 if (unlikely(!vma)) {
451 bad_area(regs, error_code, address);
452 return;
453 }
454 if (likely(vma->vm_start <= address))
455 goto good_area;
456 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
457 bad_area(regs, error_code, address);
458 return;
459 }
460 if (unlikely(expand_stack(vma, address))) {
461 bad_area(regs, error_code, address);
462 return;
463 }
464
465 /*
466 * Ok, we have a good vm_area for this memory access, so
467 * we can handle it..
468 */
469good_area:
470 if (unlikely(access_error(error_code, vma))) {
471 bad_area_access_error(regs, error_code, address);
472 return;
473 }
474
475 set_thread_fault_code(error_code);
476
477 if (user_mode(regs))
478 flags |= FAULT_FLAG_USER;
479 if (error_code & FAULT_CODE_WRITE)
480 flags |= FAULT_FLAG_WRITE;
481
482 /*
483 * If for any reason at all we couldn't handle the fault,
484 * make sure we exit gracefully rather than endlessly redo
485 * the fault.
486 */
487 fault = handle_mm_fault(mm, vma, address, flags);
488
489 if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
490 if (mm_fault_error(regs, error_code, address, fault))
491 return;
492
493 if (flags & FAULT_FLAG_ALLOW_RETRY) {
494 if (fault & VM_FAULT_MAJOR) {
495 tsk->maj_flt++;
496 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
497 regs, address);
498 } else {
499 tsk->min_flt++;
500 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
501 regs, address);
502 }
503 if (fault & VM_FAULT_RETRY) {
504 flags &= ~FAULT_FLAG_ALLOW_RETRY;
505 flags |= FAULT_FLAG_TRIED;
506
507 /*
508 * No need to up_read(&mm->mmap_sem) as we would
509 * have already released it in __lock_page_or_retry
510 * in mm/filemap.c.
511 */
512 goto retry;
513 }
514 }
515
516 up_read(&mm->mmap_sem);
517}