vserver 1.9.3
[linux-2.6.git] / arch / um / kernel / trap_kern.c
1 /* 
2  * Copyright (C) 2000, 2001 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 #include "linux/kernel.h"
7 #include "asm/errno.h"
8 #include "linux/sched.h"
9 #include "linux/mm.h"
10 #include "linux/spinlock.h"
11 #include "linux/config.h"
12 #include "linux/init.h"
13 #include "linux/ptrace.h"
14 #include "asm/semaphore.h"
15 #include "asm/pgtable.h"
16 #include "asm/tlbflush.h"
17 #include "asm/a.out.h"
18 #include "asm/current.h"
19 #include "asm/irq.h"
20 #include "user_util.h"
21 #include "kern_util.h"
22 #include "kern.h"
23 #include "chan_kern.h"
24 #include "mconsole_kern.h"
25 #include "2_5compat.h"
26 #include "mem.h"
27 #include "mem_kern.h"
28
29 int handle_page_fault(unsigned long address, unsigned long ip, 
30                       int is_write, int is_user, int *code_out)
31 {
32         struct mm_struct *mm = current->mm;
33         struct vm_area_struct *vma;
34         pgd_t *pgd;
35         pmd_t *pmd;
36         pte_t *pte;
37         unsigned long page;
38         int err = -EFAULT;
39
40         *code_out = SEGV_MAPERR;
41         down_read(&mm->mmap_sem);
42         vma = find_vma(mm, address);
43         if(!vma) 
44                 goto out;
45         else if(vma->vm_start <= address) 
46                 goto good_area;
47         else if(!(vma->vm_flags & VM_GROWSDOWN)) 
48                 goto out;
49         else if(expand_stack(vma, address)) 
50                 goto out;
51
52  good_area:
53         *code_out = SEGV_ACCERR;
54         if(is_write && !(vma->vm_flags & VM_WRITE)) 
55                 goto out;
56         page = address & PAGE_MASK;
57         pgd = pgd_offset(mm, page);
58         pmd = pmd_offset(pgd, page);
59         do {
60  survive:
61                 switch (handle_mm_fault(mm, vma, address, is_write)){
62                 case VM_FAULT_MINOR:
63                         current->min_flt++;
64                         break;
65                 case VM_FAULT_MAJOR:
66                         current->maj_flt++;
67                         break;
68                 case VM_FAULT_SIGBUS:
69                         err = -EACCES;
70                         goto out;
71                 case VM_FAULT_OOM:
72                         err = -ENOMEM;
73                         goto out_of_memory;
74                 default:
75                         BUG();
76                 }
77                 pte = pte_offset_kernel(pmd, page);
78         } while(!pte_present(*pte));
79         err = 0;
80         *pte = pte_mkyoung(*pte);
81         if(pte_write(*pte)) *pte = pte_mkdirty(*pte);
82         flush_tlb_page(vma, page);
83  out:
84         up_read(&mm->mmap_sem);
85         return(err);
86
87 /*
88  * We ran out of memory, or some other thing happened to us that made
89  * us unable to handle the page fault gracefully.
90  */
91 out_of_memory:
92         if (current->pid == 1) {
93                 up_read(&mm->mmap_sem);
94                 yield();
95                 down_read(&mm->mmap_sem);
96                 goto survive;
97         }
98         goto out;
99 }
100
101 LIST_HEAD(physmem_remappers);
102
103 void register_remapper(struct remapper *info)
104 {
105         list_add(&info->list, &physmem_remappers);
106 }
107
108 static int check_remapped_addr(unsigned long address, int is_write)
109 {
110         struct remapper *remapper;
111         struct list_head *ele;
112         __u64 offset;
113         int fd;
114
115         fd = phys_mapping(__pa(address), &offset);
116         if(fd == -1)
117                 return(0);
118
119         list_for_each(ele, &physmem_remappers){
120                 remapper = list_entry(ele, struct remapper, list);
121                 if((*remapper->proc)(fd, address, is_write, offset))
122                         return(1);
123         }
124
125         return(0);
126 }
127
128 unsigned long segv(unsigned long address, unsigned long ip, int is_write, 
129                    int is_user, void *sc)
130 {
131         struct siginfo si;
132         void *catcher;
133         int err;
134
135         if(!is_user && (address >= start_vm) && (address < end_vm)){
136                 flush_tlb_kernel_vm();
137                 return(0);
138         }
139         else if(check_remapped_addr(address & PAGE_MASK, is_write))
140                 return(0);
141         else if(current->mm == NULL)
142                 panic("Segfault with no mm");
143         err = handle_page_fault(address, ip, is_write, is_user, &si.si_code);
144
145         catcher = current->thread.fault_catcher;
146         if(!err)
147                 return(0);
148         else if(catcher != NULL){
149                 current->thread.fault_addr = (void *) address;
150                 do_longjmp(catcher, 1);
151         } 
152         else if(current->thread.fault_addr != NULL)
153                 panic("fault_addr set but no fault catcher");
154         else if(arch_fixup(ip, sc))
155                 return(0);
156
157         if(!is_user) 
158                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx", 
159                       address, ip);
160
161         if(err == -EACCES){
162                 si.si_signo = SIGBUS;
163                 si.si_errno = 0;
164                 si.si_code = BUS_ADRERR;
165                 si.si_addr = (void *)address;
166                 force_sig_info(SIGBUS, &si, current);
167         }
168         else if(err == -ENOMEM){
169                 printk("VM: killing process %s\n", current->comm);
170                 do_exit(SIGKILL);
171         }
172         else {
173                 si.si_signo = SIGSEGV;
174                 si.si_addr = (void *) address;
175                 current->thread.cr2 = address;
176                 current->thread.err = is_write;
177                 force_sig_info(SIGSEGV, &si, current);
178         }
179         return(0);
180 }
181
182 void bad_segv(unsigned long address, unsigned long ip, int is_write)
183 {
184         struct siginfo si;
185
186         si.si_signo = SIGSEGV;
187         si.si_code = SEGV_ACCERR;
188         si.si_addr = (void *) address;
189         current->thread.cr2 = address;
190         current->thread.err = is_write;
191         force_sig_info(SIGSEGV, &si, current);
192 }
193
194 void relay_signal(int sig, union uml_pt_regs *regs)
195 {
196         if(arch_handle_signal(sig, regs)) return;
197         if(!UPT_IS_USER(regs))
198                 panic("Kernel mode signal %d", sig);
199         force_sig(sig, current);
200 }
201
202 void bus_handler(int sig, union uml_pt_regs *regs)
203 {
204         if(current->thread.fault_catcher != NULL)
205                 do_longjmp(current->thread.fault_catcher, 1);
206         else relay_signal(sig, regs);
207 }
208
209 void winch(int sig, union uml_pt_regs *regs)
210 {
211         do_IRQ(WINCH_IRQ, regs);
212 }
213
214 void trap_init(void)
215 {
216 }
217
218 spinlock_t trap_lock = SPIN_LOCK_UNLOCKED;
219
220 static int trap_index = 0;
221
222 int next_trap_index(int limit)
223 {
224         int ret;
225
226         spin_lock(&trap_lock);
227         ret = trap_index;
228         if(++trap_index == limit)
229                 trap_index = 0;
230         spin_unlock(&trap_lock);
231         return(ret);
232 }
233
234 /*
235  * Overrides for Emacs so that we follow Linus's tabbing style.
236  * Emacs will notice this stuff at the end of the file and automatically
237  * adjust the settings for this buffer only.  This must remain at the end
238  * of the file.
239  * ---------------------------------------------------------------------------
240  * Local variables:
241  * c-file-style: "linux"
242  * End:
243  */