This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / arch / um / kernel / trap.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/init.h"
12 #include "linux/ptrace.h"
13 #include "asm/semaphore.h"
14 #include "asm/pgtable.h"
15 #include "asm/pgalloc.h"
16 #include "asm/tlbflush.h"
17 #include "asm/a.out.h"
18 #include "asm/current.h"
19 #include "asm/irq.h"
20 #include "sysdep/sigcontext.h"
21 #include "user_util.h"
22 #include "kern_util.h"
23 #include "kern.h"
24 #include "chan_kern.h"
25 #include "mconsole_kern.h"
26 #include "mem.h"
27 #include "mem_kern.h"
28 #include "sysdep/sigcontext.h"
29 #include "sysdep/ptrace.h"
30 #include "os.h"
31 #ifdef CONFIG_MODE_SKAS
32 #include "skas.h"
33 #endif
34 #include "os.h"
35
36 /* Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by segv(). */
37 int handle_page_fault(unsigned long address, unsigned long ip,
38                       int is_write, int is_user, int *code_out)
39 {
40         struct mm_struct *mm = current->mm;
41         struct vm_area_struct *vma;
42         pgd_t *pgd;
43         pud_t *pud;
44         pmd_t *pmd;
45         pte_t *pte;
46         int err = -EFAULT;
47
48         *code_out = SEGV_MAPERR;
49
50         /* If the fault was during atomic operation, don't take the fault, just
51          * fail. */
52         if (in_atomic())
53                 goto out_nosemaphore;
54
55         down_read(&mm->mmap_sem);
56         vma = find_vma(mm, address);
57         if(!vma)
58                 goto out;
59         else if(vma->vm_start <= address)
60                 goto good_area;
61         else if(!(vma->vm_flags & VM_GROWSDOWN))
62                 goto out;
63         else if(is_user && !ARCH_IS_STACKGROW(address))
64                 goto out;
65         else if(expand_stack(vma, address))
66                 goto out;
67
68 good_area:
69         *code_out = SEGV_ACCERR;
70         if(is_write && !(vma->vm_flags & VM_WRITE))
71                 goto out;
72
73         /* Don't require VM_READ|VM_EXEC for write faults! */
74         if(!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
75                 goto out;
76
77         do {
78 survive:
79                 switch (handle_mm_fault(mm, vma, address, is_write)){
80                 case VM_FAULT_MINOR:
81                         current->min_flt++;
82                         break;
83                 case VM_FAULT_MAJOR:
84                         current->maj_flt++;
85                         break;
86                 case VM_FAULT_SIGBUS:
87                         err = -EACCES;
88                         goto out;
89                 case VM_FAULT_OOM:
90                         err = -ENOMEM;
91                         goto out_of_memory;
92                 default:
93                         BUG();
94                 }
95                 pgd = pgd_offset(mm, address);
96                 pud = pud_offset(pgd, address);
97                 pmd = pmd_offset(pud, address);
98                 pte = pte_offset_kernel(pmd, address);
99         } while(!pte_present(*pte));
100         err = 0;
101         /* The below warning was added in place of
102          *      pte_mkyoung(); if (is_write) pte_mkdirty();
103          * If it's triggered, we'd see normally a hang here (a clean pte is
104          * marked read-only to emulate the dirty bit).
105          * However, the generic code can mark a PTE writable but clean on a
106          * concurrent read fault, triggering this harmlessly. So comment it out.
107          */
108 #if 0
109         WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
110 #endif
111         flush_tlb_page(vma, address);
112 out:
113         up_read(&mm->mmap_sem);
114 out_nosemaphore:
115         return(err);
116
117 /*
118  * We ran out of memory, or some other thing happened to us that made
119  * us unable to handle the page fault gracefully.
120  */
121 out_of_memory:
122         if (current->pid == 1) {
123                 up_read(&mm->mmap_sem);
124                 yield();
125                 down_read(&mm->mmap_sem);
126                 goto survive;
127         }
128         goto out;
129 }
130
131 void segv_handler(int sig, union uml_pt_regs *regs)
132 {
133         struct faultinfo * fi = UPT_FAULTINFO(regs);
134
135         if(UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)){
136                 bad_segv(*fi, UPT_IP(regs));
137                 return;
138         }
139         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
140 }
141
142 struct kern_handlers handlinfo_kern = {
143         .relay_signal = relay_signal,
144         .winch = winch,
145         .bus_handler = relay_signal,
146         .page_fault = segv_handler,
147         .sigio_handler = sigio_handler,
148         .timer_handler = timer_handler
149 };
150 /*
151  * We give a *copy* of the faultinfo in the regs to segv.
152  * This must be done, since nesting SEGVs could overwrite
153  * the info in the regs. A pointer to the info then would
154  * give us bad data!
155  */
156 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user, void *sc)
157 {
158         struct siginfo si;
159         void *catcher;
160         int err;
161         int is_write = FAULT_WRITE(fi);
162         unsigned long address = FAULT_ADDRESS(fi);
163
164         if(!is_user && (address >= start_vm) && (address < end_vm)){
165                 flush_tlb_kernel_vm();
166                 return(0);
167         }
168         else if(current->mm == NULL)
169                 panic("Segfault with no mm");
170
171         if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
172                 err = handle_page_fault(address, ip, is_write, is_user, &si.si_code);
173         else {
174                 err = -EFAULT;
175                 /* A thread accessed NULL, we get a fault, but CR2 is invalid.
176                  * This code is used in __do_copy_from_user() of TT mode. */
177                 address = 0;
178         }
179
180         catcher = current->thread.fault_catcher;
181         if(!err)
182                 return(0);
183         else if(catcher != NULL){
184                 current->thread.fault_addr = (void *) address;
185                 do_longjmp(catcher, 1);
186         }
187         else if(current->thread.fault_addr != NULL)
188                 panic("fault_addr set but no fault catcher");
189         else if(!is_user && arch_fixup(ip, sc))
190                 return(0);
191
192         if(!is_user)
193                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
194                       address, ip);
195
196         if (err == -EACCES) {
197                 si.si_signo = SIGBUS;
198                 si.si_errno = 0;
199                 si.si_code = BUS_ADRERR;
200                 si.si_addr = (void __user *)address;
201                 current->thread.arch.faultinfo = fi;
202                 force_sig_info(SIGBUS, &si, current);
203         } else if (err == -ENOMEM) {
204                 printk("VM: killing process %s\n", current->comm);
205                 do_exit(SIGKILL);
206         } else {
207                 BUG_ON(err != -EFAULT);
208                 si.si_signo = SIGSEGV;
209                 si.si_addr = (void __user *) address;
210                 current->thread.arch.faultinfo = fi;
211                 force_sig_info(SIGSEGV, &si, current);
212         }
213         return(0);
214 }
215
216 void bad_segv(struct faultinfo fi, unsigned long ip)
217 {
218         struct siginfo si;
219
220         si.si_signo = SIGSEGV;
221         si.si_code = SEGV_ACCERR;
222         si.si_addr = (void __user *) FAULT_ADDRESS(fi);
223         current->thread.arch.faultinfo = fi;
224         force_sig_info(SIGSEGV, &si, current);
225 }
226
227 void relay_signal(int sig, union uml_pt_regs *regs)
228 {
229         if(arch_handle_signal(sig, regs)) return;
230         if(!UPT_IS_USER(regs))
231                 panic("Kernel mode signal %d", sig);
232         current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
233         force_sig(sig, current);
234 }
235
236 void bus_handler(int sig, union uml_pt_regs *regs)
237 {
238         if(current->thread.fault_catcher != NULL)
239                 do_longjmp(current->thread.fault_catcher, 1);
240         else relay_signal(sig, regs);
241 }
242
243 void winch(int sig, union uml_pt_regs *regs)
244 {
245         do_IRQ(WINCH_IRQ, regs);
246 }
247
248 void trap_init(void)
249 {
250 }