sh: kprobes: Hook up kprobe_fault_handler() in the page fault path.
[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 <asm/io_trapped.h>
19 #include <asm/system.h>
20 #include <asm/mmu_context.h>
21 #include <asm/tlbflush.h>
22 #include <asm/kgdb.h>
23
24 #ifdef CONFIG_KPROBES
25 static inline int notify_page_fault(struct pt_regs *regs, int trap)
26 {
27         int ret = 0;
28
29         if (!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 #else
39 static inline int notify_page_fault(struct pt_regs *regs, int trap)
40 {
41         return 0;
42 }
43 #endif
44
45 /*
46  * This routine handles page faults.  It determines the address,
47  * and the problem, and then passes it off to one of the appropriate
48  * routines.
49  */
50 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
51                                         unsigned long writeaccess,
52                                         unsigned long address)
53 {
54         struct task_struct *tsk;
55         struct mm_struct *mm;
56         struct vm_area_struct * vma;
57         int si_code;
58         int fault;
59         siginfo_t info;
60
61         if (notify_page_fault(regs, writeaccess))
62                 return;
63
64 #ifdef CONFIG_SH_KGDB
65         if (kgdb_nofault && kgdb_bus_err_hook)
66                 kgdb_bus_err_hook();
67 #endif
68
69         tsk = current;
70         si_code = SEGV_MAPERR;
71
72         if (unlikely(address >= TASK_SIZE)) {
73                 /*
74                  * Synchronize this task's top level page-table
75                  * with the 'reference' page table.
76                  *
77                  * Do _not_ use "tsk" here. We might be inside
78                  * an interrupt in the middle of a task switch..
79                  */
80                 int offset = pgd_index(address);
81                 pgd_t *pgd, *pgd_k;
82                 pud_t *pud, *pud_k;
83                 pmd_t *pmd, *pmd_k;
84
85                 pgd = get_TTB() + offset;
86                 pgd_k = swapper_pg_dir + offset;
87
88                 if (!pgd_present(*pgd)) {
89                         if (!pgd_present(*pgd_k))
90                                 goto bad_area_nosemaphore;
91                         set_pgd(pgd, *pgd_k);
92                         return;
93                 }
94
95                 pud = pud_offset(pgd, address);
96                 pud_k = pud_offset(pgd_k, address);
97
98                 if (!pud_present(*pud)) {
99                         if (!pud_present(*pud_k))
100                                 goto bad_area_nosemaphore;
101                         set_pud(pud, *pud_k);
102                         return;
103                 }
104
105                 pmd = pmd_offset(pud, address);
106                 pmd_k = pmd_offset(pud_k, address);
107                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
108                         goto bad_area_nosemaphore;
109                 set_pmd(pmd, *pmd_k);
110
111                 return;
112         }
113
114         /* Only enable interrupts if they were on before the fault */
115         if ((regs->sr & SR_IMASK) != SR_IMASK) {
116                 trace_hardirqs_on();
117                 local_irq_enable();
118         }
119
120         mm = tsk->mm;
121
122         /*
123          * If we're in an interrupt or have no user
124          * context, we must not take the fault..
125          */
126         if (in_atomic() || !mm)
127                 goto no_context;
128
129         down_read(&mm->mmap_sem);
130
131         vma = find_vma(mm, address);
132         if (!vma)
133                 goto bad_area;
134         if (vma->vm_start <= address)
135                 goto good_area;
136         if (!(vma->vm_flags & VM_GROWSDOWN))
137                 goto bad_area;
138         if (expand_stack(vma, address))
139                 goto bad_area;
140 /*
141  * Ok, we have a good vm_area for this memory access, so
142  * we can handle it..
143  */
144 good_area:
145         si_code = SEGV_ACCERR;
146         if (writeaccess) {
147                 if (!(vma->vm_flags & VM_WRITE))
148                         goto bad_area;
149         } else {
150                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
151                         goto bad_area;
152         }
153
154         /*
155          * If for any reason at all we couldn't handle the fault,
156          * make sure we exit gracefully rather than endlessly redo
157          * the fault.
158          */
159 survive:
160         fault = handle_mm_fault(mm, vma, address, writeaccess);
161         if (unlikely(fault & VM_FAULT_ERROR)) {
162                 if (fault & VM_FAULT_OOM)
163                         goto out_of_memory;
164                 else if (fault & VM_FAULT_SIGBUS)
165                         goto do_sigbus;
166                 BUG();
167         }
168         if (fault & VM_FAULT_MAJOR)
169                 tsk->maj_flt++;
170         else
171                 tsk->min_flt++;
172
173         up_read(&mm->mmap_sem);
174         return;
175
176 /*
177  * Something tried to access memory that isn't in our memory map..
178  * Fix it, but check if it's kernel or user first..
179  */
180 bad_area:
181         up_read(&mm->mmap_sem);
182
183 bad_area_nosemaphore:
184         if (user_mode(regs)) {
185                 info.si_signo = SIGSEGV;
186                 info.si_errno = 0;
187                 info.si_code = si_code;
188                 info.si_addr = (void *) address;
189                 force_sig_info(SIGSEGV, &info, tsk);
190                 return;
191         }
192
193 no_context:
194         /* Are we prepared to handle this kernel fault?  */
195         if (fixup_exception(regs))
196                 return;
197
198         if (handle_trapped_io(regs, address))
199                 return;
200 /*
201  * Oops. The kernel tried to access some bad page. We'll have to
202  * terminate things with extreme prejudice.
203  *
204  */
205
206         bust_spinlocks(1);
207
208         if (oops_may_print()) {
209                 unsigned long page;
210
211                 if (address < PAGE_SIZE)
212                         printk(KERN_ALERT "Unable to handle kernel NULL "
213                                           "pointer dereference");
214                 else
215                         printk(KERN_ALERT "Unable to handle kernel paging "
216                                           "request");
217                 printk(" at virtual address %08lx\n", address);
218                 printk(KERN_ALERT "pc = %08lx\n", regs->pc);
219                 page = (unsigned long)get_TTB();
220                 if (page) {
221                         page = ((__typeof__(page) *)page)[address >> PGDIR_SHIFT];
222                         printk(KERN_ALERT "*pde = %08lx\n", page);
223                         if (page & _PAGE_PRESENT) {
224                                 page &= PAGE_MASK;
225                                 address &= 0x003ff000;
226                                 page = ((__typeof__(page) *)
227                                                 __va(page))[address >>
228                                                             PAGE_SHIFT];
229                                 printk(KERN_ALERT "*pte = %08lx\n", page);
230                         }
231                 }
232         }
233
234         die("Oops", regs, writeaccess);
235         bust_spinlocks(0);
236         do_exit(SIGKILL);
237
238 /*
239  * We ran out of memory, or some other thing happened to us that made
240  * us unable to handle the page fault gracefully.
241  */
242 out_of_memory:
243         up_read(&mm->mmap_sem);
244         if (is_global_init(current)) {
245                 yield();
246                 down_read(&mm->mmap_sem);
247                 goto survive;
248         }
249         printk("VM: killing process %s\n", tsk->comm);
250         if (user_mode(regs))
251                 do_group_exit(SIGKILL);
252         goto no_context;
253
254 do_sigbus:
255         up_read(&mm->mmap_sem);
256
257         /*
258          * Send a sigbus, regardless of whether we were in kernel
259          * or user mode.
260          */
261         info.si_signo = SIGBUS;
262         info.si_errno = 0;
263         info.si_code = BUS_ADRERR;
264         info.si_addr = (void *)address;
265         force_sig_info(SIGBUS, &info, tsk);
266
267         /* Kernel mode? Handle exceptions or die */
268         if (!user_mode(regs))
269                 goto no_context;
270 }
271
272 #ifdef CONFIG_SH_STORE_QUEUES
273 /*
274  * This is a special case for the SH-4 store queues, as pages for this
275  * space still need to be faulted in before it's possible to flush the
276  * store queue cache for writeout to the remapped region.
277  */
278 #define P3_ADDR_MAX             (P4SEG_STORE_QUE + 0x04000000)
279 #else
280 #define P3_ADDR_MAX             P4SEG
281 #endif
282
283 /*
284  * Called with interrupts disabled.
285  */
286 asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
287                                          unsigned long writeaccess,
288                                          unsigned long address)
289 {
290         pgd_t *pgd;
291         pud_t *pud;
292         pmd_t *pmd;
293         pte_t *pte;
294         pte_t entry;
295
296         if (notify_page_fault(regs, writeaccess))
297                 return 0;
298
299 #ifdef CONFIG_SH_KGDB
300         if (kgdb_nofault && kgdb_bus_err_hook)
301                 kgdb_bus_err_hook();
302 #endif
303
304         /*
305          * We don't take page faults for P1, P2, and parts of P4, these
306          * are always mapped, whether it be due to legacy behaviour in
307          * 29-bit mode, or due to PMB configuration in 32-bit mode.
308          */
309         if (address >= P3SEG && address < P3_ADDR_MAX) {
310                 pgd = pgd_offset_k(address);
311         } else {
312                 if (unlikely(address >= TASK_SIZE || !current->mm))
313                         return 1;
314
315                 pgd = pgd_offset(current->mm, address);
316         }
317
318         pud = pud_offset(pgd, address);
319         if (pud_none_or_clear_bad(pud))
320                 return 1;
321         pmd = pmd_offset(pud, address);
322         if (pmd_none_or_clear_bad(pmd))
323                 return 1;
324
325         pte = pte_offset_kernel(pmd, address);
326         entry = *pte;
327         if (unlikely(pte_none(entry) || pte_not_present(entry)))
328                 return 1;
329         if (unlikely(writeaccess && !pte_write(entry)))
330                 return 1;
331
332         if (writeaccess)
333                 entry = pte_mkdirty(entry);
334         entry = pte_mkyoung(entry);
335
336 #if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SMP)
337         /*
338          * ITLB is not affected by "ldtlb" instruction.
339          * So, we need to flush the entry by ourselves.
340          */
341         local_flush_tlb_one(get_asid(), address & PAGE_MASK);
342 #endif
343
344         set_pte(pte, entry);
345         update_mmu_cache(NULL, address, entry);
346
347         return 0;
348 }