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 if (trap == 0x300 || trap == 0x380) {
97 if (debugger_fault_handler(regs))
101 /* On a kernel SLB miss we can only check for a valid exception entry */
102 if (!user_mode(regs) && (trap == 0x380 || address >= TASK_SIZE))
105 if (error_code & 0x00400000) {
106 if (debugger_dabr_match(regs))
110 if (in_atomic() || mm == NULL) {
111 if (!user_mode(regs))
113 /* in_atomic() in user mode is really bad,
114 as is current->mm == NULL. */
115 printk(KERN_EMERG "Page fault in user mode with"
116 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
117 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
118 regs->nip, regs->msr);
119 die("Weird page fault", regs, SIGSEGV);
122 /* When running in the kernel we expect faults to occur only to
123 * addresses in user space. All other faults represent errors in the
124 * kernel and should generate an OOPS. Unfortunatly, in the case of an
125 * erroneous fault occuring in a code path which already holds mmap_sem
126 * we will deadlock attempting to validate the fault against the
127 * address space. Luckily the kernel only validly references user
128 * space from well defined areas of code, which are listed in the
131 * As the vast majority of faults will be valid we will only perform
132 * the source reference check when there is a possibilty of a deadlock.
133 * Attempt to lock the address space, if we cannot we then validate the
134 * source. If this is invalid we can skip the address space check,
135 * thus avoiding the deadlock.
137 if (!down_read_trylock(&mm->mmap_sem)) {
138 if (!user_mode(regs) && !search_exception_tables(regs->nip))
139 goto bad_area_nosemaphore;
141 down_read(&mm->mmap_sem);
144 vma = find_vma(mm, address);
148 if (vma->vm_start <= address) {
151 if (!(vma->vm_flags & VM_GROWSDOWN))
155 * N.B. The POWER/Open ABI allows programs to access up to
156 * 288 bytes below the stack pointer.
157 * The kernel signal delivery code writes up to about 1.5kB
158 * below the stack pointer (r1) before decrementing it.
159 * The exec code can write slightly over 640kB to the stack
160 * before setting the user r1. Thus we allow the stack to
161 * expand to 1MB without further checks.
163 if (address + 0x100000 < vma->vm_end) {
164 /* get user regs even if this fault is in kernel mode */
165 struct pt_regs *uregs = current->thread.regs;
170 * A user-mode access to an address a long way below
171 * the stack pointer is only valid if the instruction
172 * is one which would update the stack pointer to the
173 * address accessed if the instruction completed,
174 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
175 * (or the byte, halfword, float or double forms).
177 * If we don't check this then any write to the area
178 * between the last mapped region and the stack will
179 * expand the stack rather than segfaulting.
181 if (address + 2048 < uregs->gpr[1]
182 && (!user_mode(regs) || !store_updates_sp(regs)))
186 if (expand_stack(vma, address))
194 if (!(vma->vm_flags & VM_WRITE))
198 /* protection fault */
199 if (error_code & 0x08000000)
201 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
207 * If for any reason at all we couldn't handle the fault,
208 * make sure we exit gracefully rather than endlessly redo
211 switch (handle_mm_fault(mm, vma, address, is_write)) {
219 case VM_FAULT_SIGBUS:
227 up_read(&mm->mmap_sem);
231 up_read(&mm->mmap_sem);
233 bad_area_nosemaphore:
234 /* User mode accesses cause a SIGSEGV */
235 if (user_mode(regs)) {
236 info.si_signo = SIGSEGV;
239 info.si_addr = (void __user *) address;
240 force_sig_info(SIGSEGV, &info, current);
247 * We ran out of memory, or some other thing happened to us that made
248 * us unable to handle the page fault gracefully.
251 up_read(&mm->mmap_sem);
252 if (current->pid == 1) {
254 down_read(&mm->mmap_sem);
257 printk("VM: killing process %s\n", current->comm);
263 up_read(&mm->mmap_sem);
264 if (user_mode(regs)) {
265 info.si_signo = SIGBUS;
267 info.si_code = BUS_ADRERR;
268 info.si_addr = (void __user *)address;
269 force_sig_info(SIGBUS, &info, current);
276 * bad_page_fault is called when we have a bad access from the kernel.
277 * It is called from do_page_fault above and from some of the procedures
280 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
282 const struct exception_table_entry *entry;
284 /* Are we prepared to handle this fault? */
285 if ((entry = search_exception_tables(regs->nip)) != NULL) {
286 regs->nip = entry->fixup;
290 /* kernel has accessed a bad area */
291 die("Kernel access of bad area", regs, sig);