5 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7 * Derived from "arch/i386/mm/fault.c"
8 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
10 * Modified by Cort Dougan and Paul Mackerras.
12 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License
16 * as published by the Free Software Foundation; either version
17 * 2 of the License, or (at your option) any later version.
20 #include <linux/config.h>
21 #include <linux/signal.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/string.h>
26 #include <linux/types.h>
27 #include <linux/mman.h>
29 #include <linux/interrupt.h>
30 #include <linux/smp_lock.h>
31 #include <linux/module.h>
34 #include <asm/pgtable.h>
36 #include <asm/mmu_context.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
41 * Check whether the instruction at regs->nip is a store using
42 * an update addressing form which will update r1.
44 static int store_updates_sp(struct pt_regs
*regs
)
48 if (get_user(inst
, (unsigned int __user
*)regs
->nip
))
50 /* check for 1 in the rA field */
51 if (((inst
>> 16) & 0x1f) != 1)
53 /* check major opcode */
61 case 62: /* std or stdu */
62 return (inst
& 3) == 1;
64 /* check minor opcode */
65 switch ((inst
>> 1) & 0x3ff) {
70 case 695: /* stfsux */
71 case 759: /* stfdux */
79 * The error_code parameter is
80 * - DSISR for a non-SLB data access fault,
81 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
83 * The return value is 0 if the fault was handled, or the signal
84 * number if this is a kernel fault that can't be handled here.
86 int do_page_fault(struct pt_regs
*regs
, unsigned long address
,
87 unsigned long error_code
)
89 struct vm_area_struct
* vma
;
90 struct mm_struct
*mm
= current
->mm
;
92 unsigned long code
= SEGV_MAPERR
;
93 unsigned long is_write
= error_code
& 0x02000000;
94 unsigned long trap
= TRAP(regs
);
96 BUG_ON((trap
== 0x380) || (trap
== 0x480));
99 if (debugger_fault_handler(regs
))
103 /* On a kernel SLB miss we can only check for a valid exception entry */
104 if (!user_mode(regs
) && (address
>= TASK_SIZE
))
107 if (error_code
& 0x00400000) {
108 if (debugger_dabr_match(regs
))
112 if (in_atomic() || mm
== NULL
) {
113 if (!user_mode(regs
))
115 /* in_atomic() in user mode is really bad,
116 as is current->mm == NULL. */
117 printk(KERN_EMERG
"Page fault in user mode with"
118 "in_atomic() = %d mm = %p\n", in_atomic(), mm
);
119 printk(KERN_EMERG
"NIP = %lx MSR = %lx\n",
120 regs
->nip
, regs
->msr
);
121 die("Weird page fault", regs
, SIGSEGV
);
124 /* When running in the kernel we expect faults to occur only to
125 * addresses in user space. All other faults represent errors in the
126 * kernel and should generate an OOPS. Unfortunatly, in the case of an
127 * erroneous fault occuring in a code path which already holds mmap_sem
128 * we will deadlock attempting to validate the fault against the
129 * address space. Luckily the kernel only validly references user
130 * space from well defined areas of code, which are listed in the
133 * As the vast majority of faults will be valid we will only perform
134 * the source reference check when there is a possibilty of a deadlock.
135 * Attempt to lock the address space, if we cannot we then validate the
136 * source. If this is invalid we can skip the address space check,
137 * thus avoiding the deadlock.
139 if (!down_read_trylock(&mm
->mmap_sem
)) {
140 if (!user_mode(regs
) && !search_exception_tables(regs
->nip
))
141 goto bad_area_nosemaphore
;
143 down_read(&mm
->mmap_sem
);
146 vma
= find_vma(mm
, address
);
150 if (vma
->vm_start
<= address
) {
153 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
157 * N.B. The POWER/Open ABI allows programs to access up to
158 * 288 bytes below the stack pointer.
159 * The kernel signal delivery code writes up to about 1.5kB
160 * below the stack pointer (r1) before decrementing it.
161 * The exec code can write slightly over 640kB to the stack
162 * before setting the user r1. Thus we allow the stack to
163 * expand to 1MB without further checks.
165 if (address
+ 0x100000 < vma
->vm_end
) {
166 /* get user regs even if this fault is in kernel mode */
167 struct pt_regs
*uregs
= current
->thread
.regs
;
172 * A user-mode access to an address a long way below
173 * the stack pointer is only valid if the instruction
174 * is one which would update the stack pointer to the
175 * address accessed if the instruction completed,
176 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
177 * (or the byte, halfword, float or double forms).
179 * If we don't check this then any write to the area
180 * between the last mapped region and the stack will
181 * expand the stack rather than segfaulting.
183 if (address
+ 2048 < uregs
->gpr
[1]
184 && (!user_mode(regs
) || !store_updates_sp(regs
)))
188 if (expand_stack(vma
, address
))
196 if (!(vma
->vm_flags
& VM_WRITE
))
200 /* protection fault */
201 if (error_code
& 0x08000000)
203 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
)))
209 * If for any reason at all we couldn't handle the fault,
210 * make sure we exit gracefully rather than endlessly redo
213 switch (handle_mm_fault(mm
, vma
, address
, is_write
)) {
221 case VM_FAULT_SIGBUS
:
229 up_read(&mm
->mmap_sem
);
233 up_read(&mm
->mmap_sem
);
235 bad_area_nosemaphore
:
236 /* User mode accesses cause a SIGSEGV */
237 if (user_mode(regs
)) {
238 info
.si_signo
= SIGSEGV
;
241 info
.si_addr
= (void __user
*) address
;
242 force_sig_info(SIGSEGV
, &info
, current
);
249 * We ran out of memory, or some other thing happened to us that made
250 * us unable to handle the page fault gracefully.
253 up_read(&mm
->mmap_sem
);
254 if (current
->pid
== 1) {
256 down_read(&mm
->mmap_sem
);
259 printk("VM: killing process %s\n", current
->comm
);
265 up_read(&mm
->mmap_sem
);
266 if (user_mode(regs
)) {
267 info
.si_signo
= SIGBUS
;
269 info
.si_code
= BUS_ADRERR
;
270 info
.si_addr
= (void __user
*)address
;
271 force_sig_info(SIGBUS
, &info
, current
);
278 * bad_page_fault is called when we have a bad access from the kernel.
279 * It is called from do_page_fault above and from some of the procedures
282 void bad_page_fault(struct pt_regs
*regs
, unsigned long address
, int sig
)
284 const struct exception_table_entry
*entry
;
286 /* Are we prepared to handle this fault? */
287 if ((entry
= search_exception_tables(regs
->nip
)) != NULL
) {
288 regs
->nip
= entry
->fixup
;
292 /* kernel has accessed a bad area */
293 die("Kernel access of bad area", regs
, sig
);