Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * This file is subject to the terms and conditions of the GNU General Public | |
3 | * License. See the file "COPYING" in the main directory of this archive | |
4 | * for more details. | |
5 | * | |
6 | * Copyright (C) 1995 - 2000 by Ralf Baechle | |
7 | */ | |
8 | #include <linux/signal.h> | |
9 | #include <linux/sched.h> | |
10 | #include <linux/interrupt.h> | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/errno.h> | |
13 | #include <linux/string.h> | |
14 | #include <linux/types.h> | |
15 | #include <linux/ptrace.h> | |
16 | #include <linux/mman.h> | |
17 | #include <linux/mm.h> | |
18 | #include <linux/smp.h> | |
19 | #include <linux/smp_lock.h> | |
20 | #include <linux/vt_kern.h> /* For unblank_screen() */ | |
21 | #include <linux/module.h> | |
22 | ||
23 | #include <asm/branch.h> | |
24 | #include <asm/mmu_context.h> | |
25 | #include <asm/system.h> | |
26 | #include <asm/uaccess.h> | |
27 | #include <asm/ptrace.h> | |
28 | ||
29 | /* | |
30 | * This routine handles page faults. It determines the address, | |
31 | * and the problem, and then passes it off to one of the appropriate | |
32 | * routines. | |
33 | */ | |
34 | asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long write, | |
35 | unsigned long address) | |
36 | { | |
37 | struct vm_area_struct * vma = NULL; | |
38 | struct task_struct *tsk = current; | |
39 | struct mm_struct *mm = tsk->mm; | |
40 | const int field = sizeof(unsigned long) * 2; | |
41 | siginfo_t info; | |
42 | ||
43 | #if 0 | |
44 | printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", smp_processor_id(), | |
45 | current->comm, current->pid, field, address, write, | |
46 | field, regs->cp0_epc); | |
47 | #endif | |
48 | ||
49 | info.si_code = SEGV_MAPERR; | |
50 | ||
51 | /* | |
52 | * We fault-in kernel-space virtual memory on-demand. The | |
53 | * 'reference' page table is init_mm.pgd. | |
54 | * | |
55 | * NOTE! We MUST NOT take any locks for this case. We may | |
56 | * be in an interrupt or a critical region, and should | |
57 | * only copy the information from the master page table, | |
58 | * nothing more. | |
59 | */ | |
60 | if (unlikely(address >= VMALLOC_START)) | |
61 | goto vmalloc_fault; | |
62 | ||
63 | /* | |
64 | * If we're in an interrupt or have no user | |
65 | * context, we must not take the fault.. | |
66 | */ | |
67 | if (in_atomic() || !mm) | |
68 | goto bad_area_nosemaphore; | |
69 | ||
70 | down_read(&mm->mmap_sem); | |
71 | vma = find_vma(mm, address); | |
72 | if (!vma) | |
73 | goto bad_area; | |
74 | if (vma->vm_start <= address) | |
75 | goto good_area; | |
76 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
77 | goto bad_area; | |
78 | if (expand_stack(vma, address)) | |
79 | goto bad_area; | |
80 | /* | |
81 | * Ok, we have a good vm_area for this memory access, so | |
82 | * we can handle it.. | |
83 | */ | |
84 | good_area: | |
85 | info.si_code = SEGV_ACCERR; | |
86 | ||
87 | if (write) { | |
88 | if (!(vma->vm_flags & VM_WRITE)) | |
89 | goto bad_area; | |
90 | } else { | |
91 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) | |
92 | goto bad_area; | |
93 | } | |
94 | ||
95 | survive: | |
96 | /* | |
97 | * If for any reason at all we couldn't handle the fault, | |
98 | * make sure we exit gracefully rather than endlessly redo | |
99 | * the fault. | |
100 | */ | |
101 | switch (handle_mm_fault(mm, vma, address, write)) { | |
102 | case VM_FAULT_MINOR: | |
103 | tsk->min_flt++; | |
104 | break; | |
105 | case VM_FAULT_MAJOR: | |
106 | tsk->maj_flt++; | |
107 | break; | |
108 | case VM_FAULT_SIGBUS: | |
109 | goto do_sigbus; | |
110 | case VM_FAULT_OOM: | |
111 | goto out_of_memory; | |
112 | default: | |
113 | BUG(); | |
114 | } | |
115 | ||
116 | up_read(&mm->mmap_sem); | |
117 | return; | |
118 | ||
119 | /* | |
120 | * Something tried to access memory that isn't in our memory map.. | |
121 | * Fix it, but check if it's kernel or user first.. | |
122 | */ | |
123 | bad_area: | |
124 | up_read(&mm->mmap_sem); | |
125 | ||
126 | bad_area_nosemaphore: | |
127 | /* User mode accesses just cause a SIGSEGV */ | |
128 | if (user_mode(regs)) { | |
129 | tsk->thread.cp0_badvaddr = address; | |
130 | tsk->thread.error_code = write; | |
131 | #if 0 | |
132 | printk("do_page_fault() #2: sending SIGSEGV to %s for " | |
133 | "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n", | |
134 | tsk->comm, | |
135 | write ? "write access to" : "read access from", | |
136 | field, address, | |
137 | field, (unsigned long) regs->cp0_epc, | |
138 | field, (unsigned long) regs->regs[31]); | |
139 | #endif | |
140 | info.si_signo = SIGSEGV; | |
141 | info.si_errno = 0; | |
142 | /* info.si_code has been set above */ | |
143 | info.si_addr = (void *) address; | |
144 | force_sig_info(SIGSEGV, &info, tsk); | |
145 | return; | |
146 | } | |
147 | ||
148 | no_context: | |
149 | /* Are we prepared to handle this kernel fault? */ | |
150 | if (fixup_exception(regs)) { | |
151 | current->thread.cp0_baduaddr = address; | |
152 | return; | |
153 | } | |
154 | ||
155 | /* | |
156 | * Oops. The kernel tried to access some bad page. We'll have to | |
157 | * terminate things with extreme prejudice. | |
158 | */ | |
159 | ||
160 | bust_spinlocks(1); | |
161 | ||
162 | printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at " | |
163 | "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n", | |
164 | smp_processor_id(), field, address, field, regs->cp0_epc, | |
165 | field, regs->regs[31]); | |
166 | die("Oops", regs); | |
167 | ||
168 | /* | |
169 | * We ran out of memory, or some other thing happened to us that made | |
170 | * us unable to handle the page fault gracefully. | |
171 | */ | |
172 | out_of_memory: | |
173 | up_read(&mm->mmap_sem); | |
174 | if (tsk->pid == 1) { | |
175 | yield(); | |
176 | down_read(&mm->mmap_sem); | |
177 | goto survive; | |
178 | } | |
179 | printk("VM: killing process %s\n", tsk->comm); | |
180 | if (user_mode(regs)) | |
181 | do_exit(SIGKILL); | |
182 | goto no_context; | |
183 | ||
184 | do_sigbus: | |
185 | up_read(&mm->mmap_sem); | |
186 | ||
187 | /* Kernel mode? Handle exceptions or die */ | |
188 | if (!user_mode(regs)) | |
189 | goto no_context; | |
190 | ||
191 | /* | |
192 | * Send a sigbus, regardless of whether we were in kernel | |
193 | * or user mode. | |
194 | */ | |
195 | tsk->thread.cp0_badvaddr = address; | |
196 | info.si_signo = SIGBUS; | |
197 | info.si_errno = 0; | |
198 | info.si_code = BUS_ADRERR; | |
199 | info.si_addr = (void *) address; | |
200 | force_sig_info(SIGBUS, &info, tsk); | |
201 | ||
202 | return; | |
203 | ||
204 | vmalloc_fault: | |
205 | { | |
206 | /* | |
207 | * Synchronize this task's top level page-table | |
208 | * with the 'reference' page table. | |
209 | * | |
210 | * Do _not_ use "tsk" here. We might be inside | |
211 | * an interrupt in the middle of a task switch.. | |
212 | */ | |
213 | int offset = __pgd_offset(address); | |
214 | pgd_t *pgd, *pgd_k; | |
215 | pmd_t *pmd, *pmd_k; | |
216 | pte_t *pte_k; | |
217 | ||
218 | pgd = (pgd_t *) pgd_current[smp_processor_id()] + offset; | |
219 | pgd_k = init_mm.pgd + offset; | |
220 | ||
221 | if (!pgd_present(*pgd_k)) | |
222 | goto no_context; | |
223 | set_pgd(pgd, *pgd_k); | |
224 | ||
225 | pmd = pmd_offset(pgd, address); | |
226 | pmd_k = pmd_offset(pgd_k, address); | |
227 | if (!pmd_present(*pmd_k)) | |
228 | goto no_context; | |
229 | set_pmd(pmd, *pmd_k); | |
230 | ||
231 | pte_k = pte_offset_kernel(pmd_k, address); | |
232 | if (!pte_present(*pte_k)) | |
233 | goto no_context; | |
234 | return; | |
235 | } | |
236 | } |