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/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"
21 #include "sysdep/sigcontext.h"
22 #include "user_util.h"
23 #include "kern_util.h"
25 #include "chan_kern.h"
26 #include "mconsole_kern.h"
29 #include "sysdep/sigcontext.h"
30 #include "sysdep/ptrace.h"
32 #ifdef CONFIG_MODE_SKAS
37 /* Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by segv(). */
38 int handle_page_fault(unsigned long address
, unsigned long ip
,
39 int is_write
, int is_user
, int *code_out
)
41 struct mm_struct
*mm
= current
->mm
;
42 struct vm_area_struct
*vma
;
49 *code_out
= SEGV_MAPERR
;
51 /* If the fault was during atomic operation, don't take the fault, just
56 down_read(&mm
->mmap_sem
);
57 vma
= find_vma(mm
, address
);
60 else if(vma
->vm_start
<= address
)
62 else if(!(vma
->vm_flags
& VM_GROWSDOWN
))
64 else if(is_user
&& !ARCH_IS_STACKGROW(address
))
66 else if(expand_stack(vma
, address
))
70 *code_out
= SEGV_ACCERR
;
71 if(is_write
&& !(vma
->vm_flags
& VM_WRITE
))
74 /* Don't require VM_READ|VM_EXEC for write faults! */
75 if(!is_write
&& !(vma
->vm_flags
& (VM_READ
| VM_EXEC
)))
80 switch (handle_mm_fault(mm
, vma
, address
, is_write
)){
96 pgd
= pgd_offset(mm
, address
);
97 pud
= pud_offset(pgd
, address
);
98 pmd
= pmd_offset(pud
, address
);
99 pte
= pte_offset_kernel(pmd
, address
);
100 } while(!pte_present(*pte
));
102 /* The below warning was added in place of
103 * pte_mkyoung(); if (is_write) pte_mkdirty();
104 * If it's triggered, we'd see normally a hang here (a clean pte is
105 * marked read-only to emulate the dirty bit).
106 * However, the generic code can mark a PTE writable but clean on a
107 * concurrent read fault, triggering this harmlessly. So comment it out.
110 WARN_ON(!pte_young(*pte
) || (is_write
&& !pte_dirty(*pte
)));
112 flush_tlb_page(vma
, address
);
114 up_read(&mm
->mmap_sem
);
119 * We ran out of memory, or some other thing happened to us that made
120 * us unable to handle the page fault gracefully.
123 if (current
->pid
== 1) {
124 up_read(&mm
->mmap_sem
);
126 down_read(&mm
->mmap_sem
);
132 void segv_handler(int sig
, union uml_pt_regs
*regs
)
134 struct faultinfo
* fi
= UPT_FAULTINFO(regs
);
136 if(UPT_IS_USER(regs
) && !SEGV_IS_FIXABLE(fi
)){
137 bad_segv(*fi
, UPT_IP(regs
));
140 segv(*fi
, UPT_IP(regs
), UPT_IS_USER(regs
), regs
);
143 struct kern_handlers handlinfo_kern
= {
144 .relay_signal
= relay_signal
,
146 .bus_handler
= relay_signal
,
147 .page_fault
= segv_handler
,
148 .sigio_handler
= sigio_handler
,
149 .timer_handler
= timer_handler
152 * We give a *copy* of the faultinfo in the regs to segv.
153 * This must be done, since nesting SEGVs could overwrite
154 * the info in the regs. A pointer to the info then would
157 unsigned long segv(struct faultinfo fi
, unsigned long ip
, int is_user
, void *sc
)
162 int is_write
= FAULT_WRITE(fi
);
163 unsigned long address
= FAULT_ADDRESS(fi
);
165 if(!is_user
&& (address
>= start_vm
) && (address
< end_vm
)){
166 flush_tlb_kernel_vm();
169 else if(current
->mm
== NULL
)
170 panic("Segfault with no mm");
172 if (SEGV_IS_FIXABLE(&fi
) || SEGV_MAYBE_FIXABLE(&fi
))
173 err
= handle_page_fault(address
, ip
, is_write
, is_user
, &si
.si_code
);
176 /* A thread accessed NULL, we get a fault, but CR2 is invalid.
177 * This code is used in __do_copy_from_user() of TT mode. */
181 catcher
= current
->thread
.fault_catcher
;
184 else if(catcher
!= NULL
){
185 current
->thread
.fault_addr
= (void *) address
;
186 do_longjmp(catcher
, 1);
188 else if(current
->thread
.fault_addr
!= NULL
)
189 panic("fault_addr set but no fault catcher");
190 else if(!is_user
&& arch_fixup(ip
, sc
))
194 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
197 if (err
== -EACCES
) {
198 si
.si_signo
= SIGBUS
;
200 si
.si_code
= BUS_ADRERR
;
201 si
.si_addr
= (void __user
*)address
;
202 current
->thread
.arch
.faultinfo
= fi
;
203 force_sig_info(SIGBUS
, &si
, current
);
204 } else if (err
== -ENOMEM
) {
205 printk("VM: killing process %s\n", current
->comm
);
208 BUG_ON(err
!= -EFAULT
);
209 si
.si_signo
= SIGSEGV
;
210 si
.si_addr
= (void __user
*) address
;
211 current
->thread
.arch
.faultinfo
= fi
;
212 force_sig_info(SIGSEGV
, &si
, current
);
217 void bad_segv(struct faultinfo fi
, unsigned long ip
)
221 si
.si_signo
= SIGSEGV
;
222 si
.si_code
= SEGV_ACCERR
;
223 si
.si_addr
= (void __user
*) FAULT_ADDRESS(fi
);
224 current
->thread
.arch
.faultinfo
= fi
;
225 force_sig_info(SIGSEGV
, &si
, current
);
228 void relay_signal(int sig
, union uml_pt_regs
*regs
)
230 if(arch_handle_signal(sig
, regs
)) return;
231 if(!UPT_IS_USER(regs
))
232 panic("Kernel mode signal %d", sig
);
233 current
->thread
.arch
.faultinfo
= *UPT_FAULTINFO(regs
);
234 force_sig(sig
, current
);
237 void bus_handler(int sig
, union uml_pt_regs
*regs
)
239 if(current
->thread
.fault_catcher
!= NULL
)
240 do_longjmp(current
->thread
.fault_catcher
, 1);
241 else relay_signal(sig
, regs
);
244 void winch(int sig
, union uml_pt_regs
*regs
)
246 do_IRQ(WINCH_IRQ
, regs
);