2 * Copyright (C) 2000, 2001 Jeff Dike (jdike@karaya.com)
3 * Licensed under the GPL
6 #include "linux/kernel.h"
8 #include "linux/sched.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"
20 #include "sysdep/sigcontext.h"
21 #include "kern_util.h"
22 #include "as-layout.h"
25 #include "chan_kern.h"
26 #include "mconsole_kern.h"
29 #include "sysdep/sigcontext.h"
30 #include "sysdep/ptrace.h"
35 /* Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by segv(). */
36 int handle_page_fault(unsigned long address
, unsigned long ip
,
37 int is_write
, int is_user
, int *code_out
)
39 struct mm_struct
*mm
= current
->mm
;
40 struct vm_area_struct
*vma
;
47 *code_out
= SEGV_MAPERR
;
49 /* If the fault was during atomic operation, don't take the fault, just
54 down_read(&mm
->mmap_sem
);
55 vma
= find_vma(mm
, address
);
58 else if(vma
->vm_start
<= address
)
60 else if(!(vma
->vm_flags
& VM_GROWSDOWN
))
62 else if(is_user
&& !ARCH_IS_STACKGROW(address
))
64 else if(expand_stack(vma
, address
))
68 *code_out
= SEGV_ACCERR
;
69 if(is_write
&& !(vma
->vm_flags
& VM_WRITE
))
72 /* Don't require VM_READ|VM_EXEC for write faults! */
73 if(!is_write
&& !(vma
->vm_flags
& (VM_READ
| VM_EXEC
)))
79 fault
= handle_mm_fault(mm
, vma
, address
, is_write
);
80 if (unlikely(fault
& VM_FAULT_ERROR
)) {
81 if (fault
& VM_FAULT_OOM
) {
84 } else if (fault
& VM_FAULT_SIGBUS
) {
90 if (fault
& VM_FAULT_MAJOR
)
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
));
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.
109 WARN_ON(!pte_young(*pte
) || (is_write
&& !pte_dirty(*pte
)));
111 flush_tlb_page(vma
, address
);
113 up_read(&mm
->mmap_sem
);
118 * We ran out of memory, or some other thing happened to us that made
119 * us unable to handle the page fault gracefully.
122 if (is_init(current
)) {
123 up_read(&mm
->mmap_sem
);
125 down_read(&mm
->mmap_sem
);
131 static void bad_segv(struct faultinfo fi
, unsigned long ip
)
135 si
.si_signo
= SIGSEGV
;
136 si
.si_code
= SEGV_ACCERR
;
137 si
.si_addr
= (void __user
*) FAULT_ADDRESS(fi
);
138 current
->thread
.arch
.faultinfo
= fi
;
139 force_sig_info(SIGSEGV
, &si
, current
);
142 static void segv_handler(int sig
, union uml_pt_regs
*regs
)
144 struct faultinfo
* fi
= UPT_FAULTINFO(regs
);
146 if(UPT_IS_USER(regs
) && !SEGV_IS_FIXABLE(fi
)){
147 bad_segv(*fi
, UPT_IP(regs
));
150 segv(*fi
, UPT_IP(regs
), UPT_IS_USER(regs
), regs
);
154 * We give a *copy* of the faultinfo in the regs to segv.
155 * This must be done, since nesting SEGVs could overwrite
156 * the info in the regs. A pointer to the info then would
159 unsigned long segv(struct faultinfo fi
, unsigned long ip
, int is_user
,
160 union uml_pt_regs
*regs
)
165 int is_write
= FAULT_WRITE(fi
);
166 unsigned long address
= FAULT_ADDRESS(fi
);
168 if(!is_user
&& (address
>= start_vm
) && (address
< end_vm
)){
169 flush_tlb_kernel_vm();
172 else if(current
->mm
== NULL
) {
173 show_regs(container_of(regs
, struct pt_regs
, regs
));
174 panic("Segfault with no mm");
177 if (SEGV_IS_FIXABLE(&fi
) || SEGV_MAYBE_FIXABLE(&fi
))
178 err
= handle_page_fault(address
, ip
, is_write
, is_user
, &si
.si_code
);
181 /* A thread accessed NULL, we get a fault, but CR2 is invalid.
182 * This code is used in __do_copy_from_user() of TT mode. */
186 catcher
= current
->thread
.fault_catcher
;
189 else if(catcher
!= NULL
){
190 current
->thread
.fault_addr
= (void *) address
;
191 do_longjmp(catcher
, 1);
193 else if(current
->thread
.fault_addr
!= NULL
)
194 panic("fault_addr set but no fault catcher");
195 else if(!is_user
&& arch_fixup(ip
, regs
))
199 show_regs(container_of(regs
, struct pt_regs
, regs
));
200 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
204 if (err
== -EACCES
) {
205 si
.si_signo
= SIGBUS
;
207 si
.si_code
= BUS_ADRERR
;
208 si
.si_addr
= (void __user
*)address
;
209 current
->thread
.arch
.faultinfo
= fi
;
210 force_sig_info(SIGBUS
, &si
, current
);
211 } else if (err
== -ENOMEM
) {
212 printk("VM: killing process %s\n", current
->comm
);
215 BUG_ON(err
!= -EFAULT
);
216 si
.si_signo
= SIGSEGV
;
217 si
.si_addr
= (void __user
*) address
;
218 current
->thread
.arch
.faultinfo
= fi
;
219 force_sig_info(SIGSEGV
, &si
, current
);
224 void relay_signal(int sig
, union uml_pt_regs
*regs
)
226 if(arch_handle_signal(sig
, regs
))
229 if(!UPT_IS_USER(regs
)){
231 printk("Bus error - the host /dev/shm or /tmp mount "
232 "likely just ran out of space\n");
233 panic("Kernel mode signal %d", sig
);
236 current
->thread
.arch
.faultinfo
= *UPT_FAULTINFO(regs
);
237 force_sig(sig
, current
);
240 static void bus_handler(int sig
, union uml_pt_regs
*regs
)
242 if(current
->thread
.fault_catcher
!= NULL
)
243 do_longjmp(current
->thread
.fault_catcher
, 1);
244 else relay_signal(sig
, regs
);
247 static void winch(int sig
, union uml_pt_regs
*regs
)
249 do_IRQ(WINCH_IRQ
, regs
);
252 const struct kern_handlers handlinfo_kern
= {
253 .relay_signal
= relay_signal
,
255 .bus_handler
= bus_handler
,
256 .page_fault
= segv_handler
,
257 .sigio_handler
= sigio_handler
,
258 .timer_handler
= timer_handler