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>
32 #include <linux/kprobes.h>
35 #include <asm/pgtable.h>
37 #include <asm/mmu_context.h>
38 #include <asm/system.h>
39 #include <asm/uaccess.h>
40 #include <asm/kdebug.h>
43 * Check whether the instruction at regs->nip is a store using
44 * an update addressing form which will update r1.
46 static int store_updates_sp(struct pt_regs *regs)
50 if (get_user(inst, (unsigned int __user *)regs->nip))
52 /* check for 1 in the rA field */
53 if (((inst >> 16) & 0x1f) != 1)
55 /* check major opcode */
63 case 62: /* std or stdu */
64 return (inst & 3) == 1;
66 /* check minor opcode */
67 switch ((inst >> 1) & 0x3ff) {
72 case 695: /* stfsux */
73 case 759: /* stfdux */
80 static void do_dabr(struct pt_regs *regs, unsigned long error_code)
84 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
85 11, SIGSEGV) == NOTIFY_STOP)
88 if (debugger_dabr_match(regs))
94 /* Deliver the signal to userspace */
95 info.si_signo = SIGTRAP;
97 info.si_code = TRAP_HWBKPT;
98 info.si_addr = (void __user *)regs->nip;
99 force_sig_info(SIGTRAP, &info, current);
103 * The error_code parameter is
104 * - DSISR for a non-SLB data access fault,
105 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
107 * The return value is 0 if the fault was handled, or the signal
108 * number if this is a kernel fault that can't be handled here.
110 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
111 unsigned long error_code)
113 struct vm_area_struct * vma;
114 struct mm_struct *mm = current->mm;
116 unsigned long code = SEGV_MAPERR;
117 unsigned long is_write = error_code & DSISR_ISSTORE;
118 unsigned long trap = TRAP(regs);
119 unsigned long is_exec = trap == 0x400;
121 BUG_ON((trap == 0x380) || (trap == 0x480));
123 if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
124 11, SIGSEGV) == NOTIFY_STOP)
128 if (debugger_fault_handler(regs))
132 /* On a kernel SLB miss we can only check for a valid exception entry */
133 if (!user_mode(regs) && (address >= TASK_SIZE))
136 if (error_code & DSISR_DABRMATCH) {
137 do_dabr(regs, error_code);
141 if (in_atomic() || mm == NULL) {
142 if (!user_mode(regs))
144 /* in_atomic() in user mode is really bad,
145 as is current->mm == NULL. */
146 printk(KERN_EMERG "Page fault in user mode with"
147 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
148 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
149 regs->nip, regs->msr);
150 die("Weird page fault", regs, SIGSEGV);
153 /* When running in the kernel we expect faults to occur only to
154 * addresses in user space. All other faults represent errors in the
155 * kernel and should generate an OOPS. Unfortunatly, in the case of an
156 * erroneous fault occuring in a code path which already holds mmap_sem
157 * we will deadlock attempting to validate the fault against the
158 * address space. Luckily the kernel only validly references user
159 * space from well defined areas of code, which are listed in the
162 * As the vast majority of faults will be valid we will only perform
163 * the source reference check when there is a possibilty of a deadlock.
164 * Attempt to lock the address space, if we cannot we then validate the
165 * source. If this is invalid we can skip the address space check,
166 * thus avoiding the deadlock.
168 if (!down_read_trylock(&mm->mmap_sem)) {
169 if (!user_mode(regs) && !search_exception_tables(regs->nip))
170 goto bad_area_nosemaphore;
172 down_read(&mm->mmap_sem);
175 vma = find_vma(mm, address);
179 if (vma->vm_start <= address) {
182 if (!(vma->vm_flags & VM_GROWSDOWN))
186 * N.B. The POWER/Open ABI allows programs to access up to
187 * 288 bytes below the stack pointer.
188 * The kernel signal delivery code writes up to about 1.5kB
189 * below the stack pointer (r1) before decrementing it.
190 * The exec code can write slightly over 640kB to the stack
191 * before setting the user r1. Thus we allow the stack to
192 * expand to 1MB without further checks.
194 if (address + 0x100000 < vma->vm_end) {
195 /* get user regs even if this fault is in kernel mode */
196 struct pt_regs *uregs = current->thread.regs;
201 * A user-mode access to an address a long way below
202 * the stack pointer is only valid if the instruction
203 * is one which would update the stack pointer to the
204 * address accessed if the instruction completed,
205 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
206 * (or the byte, halfword, float or double forms).
208 * If we don't check this then any write to the area
209 * between the last mapped region and the stack will
210 * expand the stack rather than segfaulting.
212 if (address + 2048 < uregs->gpr[1]
213 && (!user_mode(regs) || !store_updates_sp(regs)))
217 if (expand_stack(vma, address))
224 /* protection fault */
225 if (error_code & DSISR_PROTFAULT)
227 if (!(vma->vm_flags & VM_EXEC))
230 } else if (is_write) {
231 if (!(vma->vm_flags & VM_WRITE))
235 if (!(vma->vm_flags & VM_READ))
241 * If for any reason at all we couldn't handle the fault,
242 * make sure we exit gracefully rather than endlessly redo
245 switch (handle_mm_fault(mm, vma, address, is_write)) {
253 case VM_FAULT_SIGBUS:
261 up_read(&mm->mmap_sem);
265 up_read(&mm->mmap_sem);
267 bad_area_nosemaphore:
268 /* User mode accesses cause a SIGSEGV */
269 if (user_mode(regs)) {
270 info.si_signo = SIGSEGV;
273 info.si_addr = (void __user *) address;
274 force_sig_info(SIGSEGV, &info, current);
278 if (trap == 0x400 && (error_code & DSISR_PROTFAULT)
279 && printk_ratelimit())
280 printk(KERN_CRIT "kernel tried to execute NX-protected"
281 " page (%lx) - exploit attempt? (uid: %d)\n",
282 address, current->uid);
287 * We ran out of memory, or some other thing happened to us that made
288 * us unable to handle the page fault gracefully.
291 up_read(&mm->mmap_sem);
292 if (current->pid == 1) {
294 down_read(&mm->mmap_sem);
297 printk("VM: killing process %s\n", current->comm);
303 up_read(&mm->mmap_sem);
304 if (user_mode(regs)) {
305 info.si_signo = SIGBUS;
307 info.si_code = BUS_ADRERR;
308 info.si_addr = (void __user *)address;
309 force_sig_info(SIGBUS, &info, current);
316 * bad_page_fault is called when we have a bad access from the kernel.
317 * It is called from do_page_fault above and from some of the procedures
320 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
322 const struct exception_table_entry *entry;
324 /* Are we prepared to handle this fault? */
325 if ((entry = search_exception_tables(regs->nip)) != NULL) {
326 regs->nip = entry->fixup;
330 /* kernel has accessed a bad area */
331 die("Kernel access of bad area", regs, sig);