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/ptrace.h>
28 #include <linux/mman.h>
30 #include <linux/interrupt.h>
31 #include <linux/highmem.h>
32 #include <linux/module.h>
33 #include <linux/kprobes.h>
36 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
41 #include <asm/tlbflush.h>
42 #include <asm/kdebug.h>
43 #include <asm/siginfo.h>
46 * Check whether the instruction at regs->nip is a store using
47 * an update addressing form which will update r1.
49 static int store_updates_sp(struct pt_regs *regs)
53 if (get_user(inst, (unsigned int __user *)regs->nip))
55 /* check for 1 in the rA field */
56 if (((inst >> 16) & 0x1f) != 1)
58 /* check major opcode */
66 case 62: /* std or stdu */
67 return (inst & 3) == 1;
69 /* check minor opcode */
70 switch ((inst >> 1) & 0x3ff) {
75 case 695: /* stfsux */
76 case 759: /* stfdux */
83 static void do_dabr(struct pt_regs *regs, unsigned long error_code)
87 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
88 11, SIGSEGV) == NOTIFY_STOP)
91 if (debugger_dabr_match(regs))
97 /* Deliver the signal to userspace */
98 info.si_signo = SIGTRAP;
100 info.si_code = TRAP_HWBKPT;
101 info.si_addr = (void __user *)regs->nip;
102 force_sig_info(SIGTRAP, &info, current);
106 * For 600- and 800-family processors, the error_code parameter is DSISR
107 * for a data fault, SRR1 for an instruction fault. For 400-family processors
108 * the error_code parameter is ESR for a data fault, 0 for an instruction
110 * For 64-bit processors, the error_code parameter is
111 * - DSISR for a non-SLB data access fault,
112 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
115 * The return value is 0 if the fault was handled, or the signal
116 * number if this is a kernel fault that can't be handled here.
118 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
119 unsigned long error_code)
121 struct vm_area_struct * vma;
122 struct mm_struct *mm = current->mm;
124 int code = SEGV_MAPERR;
126 int trap = TRAP(regs);
127 int is_exec = trap == 0x400;
129 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
131 * Fortunately the bit assignments in SRR1 for an instruction
132 * fault and DSISR for a data fault are mostly the same for the
133 * bits we are interested in. But there are some bits which
134 * indicate errors in DSISR but can validly be set in SRR1.
137 error_code &= 0x48200000;
139 is_write = error_code & DSISR_ISSTORE;
141 is_write = error_code & ESR_DST;
142 #endif /* CONFIG_4xx || CONFIG_BOOKE */
144 if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
145 11, SIGSEGV) == NOTIFY_STOP)
149 if (debugger_fault_handler(regs))
153 /* On a kernel SLB miss we can only check for a valid exception entry */
154 if (!user_mode(regs) && (address >= TASK_SIZE))
157 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
158 if (error_code & DSISR_DABRMATCH) {
160 do_dabr(regs, error_code);
163 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
165 if (in_atomic() || mm == NULL) {
166 if (!user_mode(regs))
168 /* in_atomic() in user mode is really bad,
169 as is current->mm == NULL. */
170 printk(KERN_EMERG "Page fault in user mode with"
171 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
172 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
173 regs->nip, regs->msr);
174 die("Weird page fault", regs, SIGSEGV);
177 /* When running in the kernel we expect faults to occur only to
178 * addresses in user space. All other faults represent errors in the
179 * kernel and should generate an OOPS. Unfortunatly, in the case of an
180 * erroneous fault occuring in a code path which already holds mmap_sem
181 * we will deadlock attempting to validate the fault against the
182 * address space. Luckily the kernel only validly references user
183 * space from well defined areas of code, which are listed in the
186 * As the vast majority of faults will be valid we will only perform
187 * the source reference check when there is a possibilty of a deadlock.
188 * Attempt to lock the address space, if we cannot we then validate the
189 * source. If this is invalid we can skip the address space check,
190 * thus avoiding the deadlock.
192 if (!down_read_trylock(&mm->mmap_sem)) {
193 if (!user_mode(regs) && !search_exception_tables(regs->nip))
194 goto bad_area_nosemaphore;
196 down_read(&mm->mmap_sem);
199 vma = find_vma(mm, address);
202 if (vma->vm_start <= address)
204 if (!(vma->vm_flags & VM_GROWSDOWN))
208 * N.B. The POWER/Open ABI allows programs to access up to
209 * 288 bytes below the stack pointer.
210 * The kernel signal delivery code writes up to about 1.5kB
211 * below the stack pointer (r1) before decrementing it.
212 * The exec code can write slightly over 640kB to the stack
213 * before setting the user r1. Thus we allow the stack to
214 * expand to 1MB without further checks.
216 if (address + 0x100000 < vma->vm_end) {
217 /* get user regs even if this fault is in kernel mode */
218 struct pt_regs *uregs = current->thread.regs;
223 * A user-mode access to an address a long way below
224 * the stack pointer is only valid if the instruction
225 * is one which would update the stack pointer to the
226 * address accessed if the instruction completed,
227 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
228 * (or the byte, halfword, float or double forms).
230 * If we don't check this then any write to the area
231 * between the last mapped region and the stack will
232 * expand the stack rather than segfaulting.
234 if (address + 2048 < uregs->gpr[1]
235 && (!user_mode(regs) || !store_updates_sp(regs)))
238 if (expand_stack(vma, address))
243 #if defined(CONFIG_6xx)
244 if (error_code & 0x95700000)
245 /* an error such as lwarx to I/O controller space,
246 address matching DABR, eciwx, etc. */
248 #endif /* CONFIG_6xx */
249 #if defined(CONFIG_8xx)
250 /* The MPC8xx seems to always set 0x80000000, which is
251 * "undefined". Of those that can be set, this is the only
252 * one which seems bad.
254 if (error_code & 0x10000000)
255 /* Guarded storage error. */
257 #endif /* CONFIG_8xx */
261 /* protection fault */
262 if (error_code & DSISR_PROTFAULT)
264 if (!(vma->vm_flags & VM_EXEC))
267 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
270 /* Since 4xx/Book-E supports per-page execute permission,
271 * we lazily flush dcache to icache. */
273 if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
274 struct page *page = pte_page(*ptep);
276 if (! test_bit(PG_arch_1, &page->flags)) {
277 flush_dcache_icache_page(page);
278 set_bit(PG_arch_1, &page->flags);
280 pte_update(ptep, 0, _PAGE_HWEXEC);
283 up_read(&mm->mmap_sem);
290 } else if (is_write) {
291 if (!(vma->vm_flags & VM_WRITE))
295 /* protection fault */
296 if (error_code & 0x08000000)
298 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
303 * If for any reason at all we couldn't handle the fault,
304 * make sure we exit gracefully rather than endlessly redo
308 switch (handle_mm_fault(mm, vma, address, is_write)) {
316 case VM_FAULT_SIGBUS:
324 up_read(&mm->mmap_sem);
328 up_read(&mm->mmap_sem);
330 bad_area_nosemaphore:
331 /* User mode accesses cause a SIGSEGV */
332 if (user_mode(regs)) {
333 _exception(SIGSEGV, regs, code, address);
337 if (is_exec && (error_code & DSISR_PROTFAULT)
338 && printk_ratelimit())
339 printk(KERN_CRIT "kernel tried to execute NX-protected"
340 " page (%lx) - exploit attempt? (uid: %d)\n",
341 address, current->uid);
346 * We ran out of memory, or some other thing happened to us that made
347 * us unable to handle the page fault gracefully.
350 up_read(&mm->mmap_sem);
351 if (current->pid == 1) {
353 down_read(&mm->mmap_sem);
356 printk("VM: killing process %s\n", current->comm);
362 up_read(&mm->mmap_sem);
363 if (user_mode(regs)) {
364 info.si_signo = SIGBUS;
366 info.si_code = BUS_ADRERR;
367 info.si_addr = (void __user *)address;
368 force_sig_info(SIGBUS, &info, current);
375 * bad_page_fault is called when we have a bad access from the kernel.
376 * It is called from the DSI and ISI handlers in head.S and from some
377 * of the procedures in traps.c.
379 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
381 const struct exception_table_entry *entry;
383 /* Are we prepared to handle this fault? */
384 if ((entry = search_exception_tables(regs->nip)) != NULL) {
385 regs->nip = entry->fixup;
389 /* kernel has accessed a bad area */
390 die("Kernel access of bad area", regs, sig);