sh: trapped io support V2
[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 - 2007  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 /*
25  * This routine handles page faults.  It determines the address,
26  * and the problem, and then passes it off to one of the appropriate
27  * routines.
28  */
29 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
30                                         unsigned long writeaccess,
31                                         unsigned long address)
32 {
33         struct task_struct *tsk;
34         struct mm_struct *mm;
35         struct vm_area_struct * vma;
36         int si_code;
37         int fault;
38         siginfo_t info;
39
40         trace_hardirqs_on();
41         local_irq_enable();
42
43 #ifdef CONFIG_SH_KGDB
44         if (kgdb_nofault && kgdb_bus_err_hook)
45                 kgdb_bus_err_hook();
46 #endif
47
48         tsk = current;
49         mm = tsk->mm;
50         si_code = SEGV_MAPERR;
51
52         if (unlikely(address >= TASK_SIZE)) {
53                 /*
54                  * Synchronize this task's top level page-table
55                  * with the 'reference' page table.
56                  *
57                  * Do _not_ use "tsk" here. We might be inside
58                  * an interrupt in the middle of a task switch..
59                  */
60                 int offset = pgd_index(address);
61                 pgd_t *pgd, *pgd_k;
62                 pud_t *pud, *pud_k;
63                 pmd_t *pmd, *pmd_k;
64
65                 pgd = get_TTB() + offset;
66                 pgd_k = swapper_pg_dir + offset;
67
68                 /* This will never happen with the folded page table. */
69                 if (!pgd_present(*pgd)) {
70                         if (!pgd_present(*pgd_k))
71                                 goto bad_area_nosemaphore;
72                         set_pgd(pgd, *pgd_k);
73                         return;
74                 }
75
76                 pud = pud_offset(pgd, address);
77                 pud_k = pud_offset(pgd_k, address);
78                 if (pud_present(*pud) || !pud_present(*pud_k))
79                         goto bad_area_nosemaphore;
80                 set_pud(pud, *pud_k);
81
82                 pmd = pmd_offset(pud, address);
83                 pmd_k = pmd_offset(pud_k, address);
84                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
85                         goto bad_area_nosemaphore;
86                 set_pmd(pmd, *pmd_k);
87
88                 return;
89         }
90
91         /*
92          * If we're in an interrupt or have no user
93          * context, we must not take the fault..
94          */
95         if (in_atomic() || !mm)
96                 goto no_context;
97
98         down_read(&mm->mmap_sem);
99
100         vma = find_vma(mm, address);
101         if (!vma)
102                 goto bad_area;
103         if (vma->vm_start <= address)
104                 goto good_area;
105         if (!(vma->vm_flags & VM_GROWSDOWN))
106                 goto bad_area;
107         if (expand_stack(vma, address))
108                 goto bad_area;
109 /*
110  * Ok, we have a good vm_area for this memory access, so
111  * we can handle it..
112  */
113 good_area:
114         si_code = SEGV_ACCERR;
115         if (writeaccess) {
116                 if (!(vma->vm_flags & VM_WRITE))
117                         goto bad_area;
118         } else {
119                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
120                         goto bad_area;
121         }
122
123         /*
124          * If for any reason at all we couldn't handle the fault,
125          * make sure we exit gracefully rather than endlessly redo
126          * the fault.
127          */
128 survive:
129         fault = handle_mm_fault(mm, vma, address, writeaccess);
130         if (unlikely(fault & VM_FAULT_ERROR)) {
131                 if (fault & VM_FAULT_OOM)
132                         goto out_of_memory;
133                 else if (fault & VM_FAULT_SIGBUS)
134                         goto do_sigbus;
135                 BUG();
136         }
137         if (fault & VM_FAULT_MAJOR)
138                 tsk->maj_flt++;
139         else
140                 tsk->min_flt++;
141
142         up_read(&mm->mmap_sem);
143         return;
144
145 /*
146  * Something tried to access memory that isn't in our memory map..
147  * Fix it, but check if it's kernel or user first..
148  */
149 bad_area:
150         up_read(&mm->mmap_sem);
151
152 bad_area_nosemaphore:
153         if (user_mode(regs)) {
154                 info.si_signo = SIGSEGV;
155                 info.si_errno = 0;
156                 info.si_code = si_code;
157                 info.si_addr = (void *) address;
158                 force_sig_info(SIGSEGV, &info, tsk);
159                 return;
160         }
161
162 no_context:
163         /* Are we prepared to handle this kernel fault?  */
164         if (fixup_exception(regs))
165                 return;
166
167         if (handle_trapped_io(regs, address))
168                 return;
169 /*
170  * Oops. The kernel tried to access some bad page. We'll have to
171  * terminate things with extreme prejudice.
172  *
173  */
174
175         bust_spinlocks(1);
176
177         if (oops_may_print()) {
178                 unsigned long page;
179
180                 if (address < PAGE_SIZE)
181                         printk(KERN_ALERT "Unable to handle kernel NULL "
182                                           "pointer dereference");
183                 else
184                         printk(KERN_ALERT "Unable to handle kernel paging "
185                                           "request");
186                 printk(" at virtual address %08lx\n", address);
187                 printk(KERN_ALERT "pc = %08lx\n", regs->pc);
188                 page = (unsigned long)get_TTB();
189                 if (page) {
190                         page = ((__typeof__(page) *)page)[address >> PGDIR_SHIFT];
191                         printk(KERN_ALERT "*pde = %08lx\n", page);
192                         if (page & _PAGE_PRESENT) {
193                                 page &= PAGE_MASK;
194                                 address &= 0x003ff000;
195                                 page = ((__typeof__(page) *)
196                                                 __va(page))[address >>
197                                                             PAGE_SHIFT];
198                                 printk(KERN_ALERT "*pte = %08lx\n", page);
199                         }
200                 }
201         }
202
203         die("Oops", regs, writeaccess);
204         bust_spinlocks(0);
205         do_exit(SIGKILL);
206
207 /*
208  * We ran out of memory, or some other thing happened to us that made
209  * us unable to handle the page fault gracefully.
210  */
211 out_of_memory:
212         up_read(&mm->mmap_sem);
213         if (is_global_init(current)) {
214                 yield();
215                 down_read(&mm->mmap_sem);
216                 goto survive;
217         }
218         printk("VM: killing process %s\n", tsk->comm);
219         if (user_mode(regs))
220                 do_group_exit(SIGKILL);
221         goto no_context;
222
223 do_sigbus:
224         up_read(&mm->mmap_sem);
225
226         /*
227          * Send a sigbus, regardless of whether we were in kernel
228          * or user mode.
229          */
230         info.si_signo = SIGBUS;
231         info.si_errno = 0;
232         info.si_code = BUS_ADRERR;
233         info.si_addr = (void *)address;
234         force_sig_info(SIGBUS, &info, tsk);
235
236         /* Kernel mode? Handle exceptions or die */
237         if (!user_mode(regs))
238                 goto no_context;
239 }
240
241 #ifdef CONFIG_SH_STORE_QUEUES
242 /*
243  * This is a special case for the SH-4 store queues, as pages for this
244  * space still need to be faulted in before it's possible to flush the
245  * store queue cache for writeout to the remapped region.
246  */
247 #define P3_ADDR_MAX             (P4SEG_STORE_QUE + 0x04000000)
248 #else
249 #define P3_ADDR_MAX             P4SEG
250 #endif
251
252 /*
253  * Called with interrupts disabled.
254  */
255 asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
256                                          unsigned long writeaccess,
257                                          unsigned long address)
258 {
259         pgd_t *pgd;
260         pud_t *pud;
261         pmd_t *pmd;
262         pte_t *pte;
263         pte_t entry;
264
265 #ifdef CONFIG_SH_KGDB
266         if (kgdb_nofault && kgdb_bus_err_hook)
267                 kgdb_bus_err_hook();
268 #endif
269
270         /*
271          * We don't take page faults for P1, P2, and parts of P4, these
272          * are always mapped, whether it be due to legacy behaviour in
273          * 29-bit mode, or due to PMB configuration in 32-bit mode.
274          */
275         if (address >= P3SEG && address < P3_ADDR_MAX) {
276                 pgd = pgd_offset_k(address);
277         } else {
278                 if (unlikely(address >= TASK_SIZE || !current->mm))
279                         return 1;
280
281                 pgd = pgd_offset(current->mm, address);
282         }
283
284         pud = pud_offset(pgd, address);
285         if (pud_none_or_clear_bad(pud))
286                 return 1;
287         pmd = pmd_offset(pud, address);
288         if (pmd_none_or_clear_bad(pmd))
289                 return 1;
290
291         pte = pte_offset_kernel(pmd, address);
292         entry = *pte;
293         if (unlikely(pte_none(entry) || pte_not_present(entry)))
294                 return 1;
295         if (unlikely(writeaccess && !pte_write(entry)))
296                 return 1;
297
298         if (writeaccess)
299                 entry = pte_mkdirty(entry);
300         entry = pte_mkyoung(entry);
301
302         set_pte(pte, entry);
303         update_mmu_cache(NULL, address, entry);
304
305         return 0;
306 }