sh: use kprobes_built_in() for notify_page_fault().
[linux-2.6.git] / arch / sh / mm / fault_32.c
1 /*
2  * Page fault handler for SH with an MMU.
3  *
4  *  Copyright (C) 1999  Niibe Yutaka
5  *  Copyright (C) 2003 - 2008  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_counter.h>
19 #include <asm/io_trapped.h>
20 #include <asm/system.h>
21 #include <asm/mmu_context.h>
22 #include <asm/tlbflush.h>
23
24 static inline int notify_page_fault(struct pt_regs *regs, int trap)
25 {
26         int ret = 0;
27
28         if (kprobes_built_in() && !user_mode(regs)) {
29                 preempt_disable();
30                 if (kprobe_running() && kprobe_fault_handler(regs, trap))
31                         ret = 1;
32                 preempt_enable();
33         }
34
35         return ret;
36 }
37
38 /*
39  * This routine handles page faults.  It determines the address,
40  * and the problem, and then passes it off to one of the appropriate
41  * routines.
42  */
43 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
44                                         unsigned long writeaccess,
45                                         unsigned long address)
46 {
47         struct task_struct *tsk;
48         struct mm_struct *mm;
49         struct vm_area_struct * vma;
50         int si_code;
51         int fault;
52         siginfo_t info;
53
54         /*
55          * We don't bother with any notifier callbacks here, as they are
56          * all handled through the __do_page_fault() fast-path.
57          */
58
59         tsk = current;
60         si_code = SEGV_MAPERR;
61
62         if (unlikely(address >= TASK_SIZE)) {
63                 /*
64                  * Synchronize this task's top level page-table
65                  * with the 'reference' page table.
66                  *
67                  * Do _not_ use "tsk" here. We might be inside
68                  * an interrupt in the middle of a task switch..
69                  */
70                 int offset = pgd_index(address);
71                 pgd_t *pgd, *pgd_k;
72                 pud_t *pud, *pud_k;
73                 pmd_t *pmd, *pmd_k;
74
75                 pgd = get_TTB() + offset;
76                 pgd_k = swapper_pg_dir + offset;
77
78                 if (!pgd_present(*pgd)) {
79                         if (!pgd_present(*pgd_k))
80                                 goto bad_area_nosemaphore;
81                         set_pgd(pgd, *pgd_k);
82                         return;
83                 }
84
85                 pud = pud_offset(pgd, address);
86                 pud_k = pud_offset(pgd_k, address);
87
88                 if (!pud_present(*pud)) {
89                         if (!pud_present(*pud_k))
90                                 goto bad_area_nosemaphore;
91                         set_pud(pud, *pud_k);
92                         return;
93                 }
94
95                 pmd = pmd_offset(pud, address);
96                 pmd_k = pmd_offset(pud_k, address);
97                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
98                         goto bad_area_nosemaphore;
99                 set_pmd(pmd, *pmd_k);
100
101                 return;
102         }
103
104         mm = tsk->mm;
105
106         if (unlikely(notify_page_fault(regs, lookup_exception_vector())))
107                 return;
108
109         /* Only enable interrupts if they were on before the fault */
110         if ((regs->sr & SR_IMASK) != SR_IMASK)
111                 local_irq_enable();
112
113         perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
114
115         /*
116          * If we're in an interrupt or have no user
117          * context, we must not take the fault..
118          */
119         if (in_atomic() || !mm)
120                 goto no_context;
121
122         down_read(&mm->mmap_sem);
123
124         vma = find_vma(mm, address);
125         if (!vma)
126                 goto bad_area;
127         if (vma->vm_start <= address)
128                 goto good_area;
129         if (!(vma->vm_flags & VM_GROWSDOWN))
130                 goto bad_area;
131         if (expand_stack(vma, address))
132                 goto bad_area;
133 /*
134  * Ok, we have a good vm_area for this memory access, so
135  * we can handle it..
136  */
137 good_area:
138         si_code = SEGV_ACCERR;
139         if (writeaccess) {
140                 if (!(vma->vm_flags & VM_WRITE))
141                         goto bad_area;
142         } else {
143                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
144                         goto bad_area;
145         }
146
147         /*
148          * If for any reason at all we couldn't handle the fault,
149          * make sure we exit gracefully rather than endlessly redo
150          * the fault.
151          */
152 survive:
153         fault = handle_mm_fault(mm, vma, address, writeaccess ? FAULT_FLAG_WRITE : 0);
154         if (unlikely(fault & VM_FAULT_ERROR)) {
155                 if (fault & VM_FAULT_OOM)
156                         goto out_of_memory;
157                 else if (fault & VM_FAULT_SIGBUS)
158                         goto do_sigbus;
159                 BUG();
160         }
161         if (fault & VM_FAULT_MAJOR) {
162                 tsk->maj_flt++;
163                 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
164                                      regs, address);
165         } else {
166                 tsk->min_flt++;
167                 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
168                                      regs, address);
169         }
170
171         up_read(&mm->mmap_sem);
172         return;
173
174 /*
175  * Something tried to access memory that isn't in our memory map..
176  * Fix it, but check if it's kernel or user first..
177  */
178 bad_area:
179         up_read(&mm->mmap_sem);
180
181 bad_area_nosemaphore:
182         if (user_mode(regs)) {
183                 info.si_signo = SIGSEGV;
184                 info.si_errno = 0;
185                 info.si_code = si_code;
186                 info.si_addr = (void *) address;
187                 force_sig_info(SIGSEGV, &info, tsk);
188                 return;
189         }
190
191 no_context:
192         /* Are we prepared to handle this kernel fault?  */
193         if (fixup_exception(regs))
194                 return;
195
196         if (handle_trapped_io(regs, address))
197                 return;
198 /*
199  * Oops. The kernel tried to access some bad page. We'll have to
200  * terminate things with extreme prejudice.
201  *
202  */
203
204         bust_spinlocks(1);
205
206         if (oops_may_print()) {
207                 unsigned long page;
208
209                 if (address < PAGE_SIZE)
210                         printk(KERN_ALERT "Unable to handle kernel NULL "
211                                           "pointer dereference");
212                 else
213                         printk(KERN_ALERT "Unable to handle kernel paging "
214                                           "request");
215                 printk(" at virtual address %08lx\n", address);
216                 printk(KERN_ALERT "pc = %08lx\n", regs->pc);
217                 page = (unsigned long)get_TTB();
218                 if (page) {
219                         page = ((__typeof__(page) *)page)[address >> PGDIR_SHIFT];
220                         printk(KERN_ALERT "*pde = %08lx\n", page);
221                         if (page & _PAGE_PRESENT) {
222                                 page &= PAGE_MASK;
223                                 address &= 0x003ff000;
224                                 page = ((__typeof__(page) *)
225                                                 __va(page))[address >>
226                                                             PAGE_SHIFT];
227                                 printk(KERN_ALERT "*pte = %08lx\n", page);
228                         }
229                 }
230         }
231
232         die("Oops", regs, writeaccess);
233         bust_spinlocks(0);
234         do_exit(SIGKILL);
235
236 /*
237  * We ran out of memory, or some other thing happened to us that made
238  * us unable to handle the page fault gracefully.
239  */
240 out_of_memory:
241         up_read(&mm->mmap_sem);
242         if (is_global_init(current)) {
243                 yield();
244                 down_read(&mm->mmap_sem);
245                 goto survive;
246         }
247         printk("VM: killing process %s\n", tsk->comm);
248         if (user_mode(regs))
249                 do_group_exit(SIGKILL);
250         goto no_context;
251
252 do_sigbus:
253         up_read(&mm->mmap_sem);
254
255         /*
256          * Send a sigbus, regardless of whether we were in kernel
257          * or user mode.
258          */
259         info.si_signo = SIGBUS;
260         info.si_errno = 0;
261         info.si_code = BUS_ADRERR;
262         info.si_addr = (void *)address;
263         force_sig_info(SIGBUS, &info, tsk);
264
265         /* Kernel mode? Handle exceptions or die */
266         if (!user_mode(regs))
267                 goto no_context;
268 }
269
270 /*
271  * Called with interrupts disabled.
272  */
273 asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
274                                          unsigned long writeaccess,
275                                          unsigned long address)
276 {
277         pgd_t *pgd;
278         pud_t *pud;
279         pmd_t *pmd;
280         pte_t *pte;
281         pte_t entry;
282         int ret = 1;
283
284         /*
285          * We don't take page faults for P1, P2, and parts of P4, these
286          * are always mapped, whether it be due to legacy behaviour in
287          * 29-bit mode, or due to PMB configuration in 32-bit mode.
288          */
289         if (address >= P3SEG && address < P3_ADDR_MAX) {
290                 pgd = pgd_offset_k(address);
291         } else {
292                 if (unlikely(address >= TASK_SIZE || !current->mm))
293                         goto out;
294
295                 pgd = pgd_offset(current->mm, address);
296         }
297
298         pud = pud_offset(pgd, address);
299         if (pud_none_or_clear_bad(pud))
300                 goto out;
301         pmd = pmd_offset(pud, address);
302         if (pmd_none_or_clear_bad(pmd))
303                 goto out;
304         pte = pte_offset_kernel(pmd, address);
305         entry = *pte;
306         if (unlikely(pte_none(entry) || pte_not_present(entry)))
307                 goto out;
308         if (unlikely(writeaccess && !pte_write(entry)))
309                 goto out;
310
311         if (writeaccess)
312                 entry = pte_mkdirty(entry);
313         entry = pte_mkyoung(entry);
314
315 #if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SMP)
316         /*
317          * ITLB is not affected by "ldtlb" instruction.
318          * So, we need to flush the entry by ourselves.
319          */
320         local_flush_tlb_one(get_asid(), address & PAGE_MASK);
321 #endif
322
323         set_pte(pte, entry);
324         update_mmu_cache(NULL, address, entry);
325
326         ret = 0;
327 out:
328         return ret;
329 }