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/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}