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