Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid
[linux-2.6-microblaze.git] / arch / sh / mm / fault.c
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
27 static void
28 force_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  */
37 static 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
120 static 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  */
176 static 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
205 static void
206 show_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
221 static noinline void
222 no_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         bust_spinlocks(0);
242         do_exit(SIGKILL);
243 }
244
245 static void
246 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
247                        unsigned long address, int si_code)
248 {
249         /* User mode accesses just cause a SIGSEGV */
250         if (user_mode(regs)) {
251                 /*
252                  * It's possible to have interrupts off here:
253                  */
254                 local_irq_enable();
255
256                 force_sig_info_fault(SIGSEGV, si_code, address);
257
258                 return;
259         }
260
261         no_context(regs, error_code, address);
262 }
263
264 static noinline void
265 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
266                      unsigned long address)
267 {
268         __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
269 }
270
271 static void
272 __bad_area(struct pt_regs *regs, unsigned long error_code,
273            unsigned long address, int si_code)
274 {
275         struct mm_struct *mm = current->mm;
276
277         /*
278          * Something tried to access memory that isn't in our memory map..
279          * Fix it, but check if it's kernel or user first..
280          */
281         mmap_read_unlock(mm);
282
283         __bad_area_nosemaphore(regs, error_code, address, si_code);
284 }
285
286 static noinline void
287 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
288 {
289         __bad_area(regs, error_code, address, SEGV_MAPERR);
290 }
291
292 static noinline void
293 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
294                       unsigned long address)
295 {
296         __bad_area(regs, error_code, address, SEGV_ACCERR);
297 }
298
299 static void
300 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
301 {
302         struct task_struct *tsk = current;
303         struct mm_struct *mm = tsk->mm;
304
305         mmap_read_unlock(mm);
306
307         /* Kernel mode? Handle exceptions or die: */
308         if (!user_mode(regs))
309                 no_context(regs, error_code, address);
310
311         force_sig_info_fault(SIGBUS, BUS_ADRERR, address);
312 }
313
314 static noinline int
315 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
316                unsigned long address, vm_fault_t fault)
317 {
318         /*
319          * Pagefault was interrupted by SIGKILL. We have no reason to
320          * continue pagefault.
321          */
322         if (fault_signal_pending(fault, regs)) {
323                 if (!user_mode(regs))
324                         no_context(regs, error_code, address);
325                 return 1;
326         }
327
328         /* Release mmap_lock first if necessary */
329         if (!(fault & VM_FAULT_RETRY))
330                 mmap_read_unlock(current->mm);
331
332         if (!(fault & VM_FAULT_ERROR))
333                 return 0;
334
335         if (fault & VM_FAULT_OOM) {
336                 /* Kernel mode? Handle exceptions or die: */
337                 if (!user_mode(regs)) {
338                         no_context(regs, error_code, address);
339                         return 1;
340                 }
341
342                 /*
343                  * We ran out of memory, call the OOM killer, and return the
344                  * userspace (which will retry the fault, or kill us if we got
345                  * oom-killed):
346                  */
347                 pagefault_out_of_memory();
348         } else {
349                 if (fault & VM_FAULT_SIGBUS)
350                         do_sigbus(regs, error_code, address);
351                 else if (fault & VM_FAULT_SIGSEGV)
352                         bad_area(regs, error_code, address);
353                 else
354                         BUG();
355         }
356
357         return 1;
358 }
359
360 static inline int access_error(int error_code, struct vm_area_struct *vma)
361 {
362         if (error_code & FAULT_CODE_WRITE) {
363                 /* write, present and write, not present: */
364                 if (unlikely(!(vma->vm_flags & VM_WRITE)))
365                         return 1;
366                 return 0;
367         }
368
369         /* ITLB miss on NX page */
370         if (unlikely((error_code & FAULT_CODE_ITLB) &&
371                      !(vma->vm_flags & VM_EXEC)))
372                 return 1;
373
374         /* read, not present: */
375         if (unlikely(!vma_is_accessible(vma)))
376                 return 1;
377
378         return 0;
379 }
380
381 static int fault_in_kernel_space(unsigned long address)
382 {
383         return address >= TASK_SIZE;
384 }
385
386 /*
387  * This routine handles page faults.  It determines the address,
388  * and the problem, and then passes it off to one of the appropriate
389  * routines.
390  */
391 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
392                                         unsigned long error_code,
393                                         unsigned long address)
394 {
395         unsigned long vec;
396         struct task_struct *tsk;
397         struct mm_struct *mm;
398         struct vm_area_struct * vma;
399         vm_fault_t fault;
400         unsigned int flags = FAULT_FLAG_DEFAULT;
401
402         tsk = current;
403         mm = tsk->mm;
404         vec = lookup_exception_vector();
405
406         /*
407          * We fault-in kernel-space virtual memory on-demand. The
408          * 'reference' page table is init_mm.pgd.
409          *
410          * NOTE! We MUST NOT take any locks for this case. We may
411          * be in an interrupt or a critical region, and should
412          * only copy the information from the master page table,
413          * nothing more.
414          */
415         if (unlikely(fault_in_kernel_space(address))) {
416                 if (vmalloc_fault(address) >= 0)
417                         return;
418                 if (kprobe_page_fault(regs, vec))
419                         return;
420
421                 bad_area_nosemaphore(regs, error_code, address);
422                 return;
423         }
424
425         if (unlikely(kprobe_page_fault(regs, vec)))
426                 return;
427
428         /* Only enable interrupts if they were on before the fault */
429         if ((regs->sr & SR_IMASK) != SR_IMASK)
430                 local_irq_enable();
431
432         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
433
434         /*
435          * If we're in an interrupt, have no user context or are running
436          * with pagefaults disabled then we must not take the fault:
437          */
438         if (unlikely(faulthandler_disabled() || !mm)) {
439                 bad_area_nosemaphore(regs, error_code, address);
440                 return;
441         }
442
443 retry:
444         mmap_read_lock(mm);
445
446         vma = find_vma(mm, address);
447         if (unlikely(!vma)) {
448                 bad_area(regs, error_code, address);
449                 return;
450         }
451         if (likely(vma->vm_start <= address))
452                 goto good_area;
453         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
454                 bad_area(regs, error_code, address);
455                 return;
456         }
457         if (unlikely(expand_stack(vma, address))) {
458                 bad_area(regs, error_code, address);
459                 return;
460         }
461
462         /*
463          * Ok, we have a good vm_area for this memory access, so
464          * we can handle it..
465          */
466 good_area:
467         if (unlikely(access_error(error_code, vma))) {
468                 bad_area_access_error(regs, error_code, address);
469                 return;
470         }
471
472         set_thread_fault_code(error_code);
473
474         if (user_mode(regs))
475                 flags |= FAULT_FLAG_USER;
476         if (error_code & FAULT_CODE_WRITE)
477                 flags |= FAULT_FLAG_WRITE;
478
479         /*
480          * If for any reason at all we couldn't handle the fault,
481          * make sure we exit gracefully rather than endlessly redo
482          * the fault.
483          */
484         fault = handle_mm_fault(vma, address, flags, regs);
485
486         if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
487                 if (mm_fault_error(regs, error_code, address, fault))
488                         return;
489
490         if (flags & FAULT_FLAG_ALLOW_RETRY) {
491                 if (fault & VM_FAULT_RETRY) {
492                         flags |= FAULT_FLAG_TRIED;
493
494                         /*
495                          * No need to mmap_read_unlock(mm) as we would
496                          * have already released it in __lock_page_or_retry
497                          * in mm/filemap.c.
498                          */
499                         goto retry;
500                 }
501         }
502
503         mmap_read_unlock(mm);
504 }