3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5 * Derived from "arch/i386/mm/fault.c"
6 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Modified by Cort Dougan and Paul Mackerras.
10 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
18 #include <linux/config.h>
19 #include <linux/signal.h>
20 #include <linux/sched.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/string.h>
24 #include <linux/types.h>
25 #include <linux/ptrace.h>
26 #include <linux/mman.h>
28 #include <linux/interrupt.h>
29 #include <linux/highmem.h>
30 #include <linux/module.h>
31 #include <linux/kprobes.h>
34 #include <asm/pgtable.h>
36 #include <asm/mmu_context.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/tlbflush.h>
40 #include <asm/kdebug.h>
41 #include <asm/siginfo.h>
44 * Check whether the instruction at regs->nip is a store using
45 * an update addressing form which will update r1.
47 static int store_updates_sp(struct pt_regs *regs)
51 if (get_user(inst, (unsigned int __user *)regs->nip))
53 /* check for 1 in the rA field */
54 if (((inst >> 16) & 0x1f) != 1)
56 /* check major opcode */
64 case 62: /* std or stdu */
65 return (inst & 3) == 1;
67 /* check minor opcode */
68 switch ((inst >> 1) & 0x3ff) {
73 case 695: /* stfsux */
74 case 759: /* stfdux */
81 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
82 static void do_dabr(struct pt_regs *regs, unsigned long address,
83 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 *)address;
102 force_sig_info(SIGTRAP, &info, current);
104 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
107 * For 600- and 800-family processors, the error_code parameter is DSISR
108 * for a data fault, SRR1 for an instruction fault. For 400-family processors
109 * the error_code parameter is ESR for a data fault, 0 for an instruction
111 * For 64-bit processors, the error_code parameter is
112 * - DSISR for a non-SLB data access fault,
113 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
116 * The return value is 0 if the fault was handled, or the signal
117 * number if this is a kernel fault that can't be handled here.
119 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
120 unsigned long error_code)
122 struct vm_area_struct * vma;
123 struct mm_struct *mm = current->mm;
125 int code = SEGV_MAPERR;
127 int trap = TRAP(regs);
128 int is_exec = trap == 0x400;
130 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
132 * Fortunately the bit assignments in SRR1 for an instruction
133 * fault and DSISR for a data fault are mostly the same for the
134 * bits we are interested in. But there are some bits which
135 * indicate errors in DSISR but can validly be set in SRR1.
138 error_code &= 0x48200000;
140 is_write = error_code & DSISR_ISSTORE;
142 is_write = error_code & ESR_DST;
143 #endif /* CONFIG_4xx || CONFIG_BOOKE */
145 if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
146 11, SIGSEGV) == NOTIFY_STOP)
150 if (debugger_fault_handler(regs))
154 /* On a kernel SLB miss we can only check for a valid exception entry */
155 if (!user_mode(regs) && (address >= TASK_SIZE))
158 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
159 if (error_code & DSISR_DABRMATCH) {
161 do_dabr(regs, address, error_code);
164 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
166 if (in_atomic() || mm == NULL) {
167 if (!user_mode(regs))
169 /* in_atomic() in user mode is really bad,
170 as is current->mm == NULL. */
171 printk(KERN_EMERG "Page fault in user mode with"
172 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
173 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
174 regs->nip, regs->msr);
175 die("Weird page fault", regs, SIGSEGV);
178 /* When running in the kernel we expect faults to occur only to
179 * addresses in user space. All other faults represent errors in the
180 * kernel and should generate an OOPS. Unfortunately, in the case of an
181 * erroneous fault occurring in a code path which already holds mmap_sem
182 * we will deadlock attempting to validate the fault against the
183 * address space. Luckily the kernel only validly references user
184 * space from well defined areas of code, which are listed in the
187 * As the vast majority of faults will be valid we will only perform
188 * the source reference check when there is a possibility of a deadlock.
189 * Attempt to lock the address space, if we cannot we then validate the
190 * source. If this is invalid we can skip the address space check,
191 * thus avoiding the deadlock.
193 if (!down_read_trylock(&mm->mmap_sem)) {
194 if (!user_mode(regs) && !search_exception_tables(regs->nip))
195 goto bad_area_nosemaphore;
197 down_read(&mm->mmap_sem);
200 vma = find_vma(mm, address);
203 if (vma->vm_start <= address)
205 if (!(vma->vm_flags & VM_GROWSDOWN))
209 * N.B. The POWER/Open ABI allows programs to access up to
210 * 288 bytes below the stack pointer.
211 * The kernel signal delivery code writes up to about 1.5kB
212 * below the stack pointer (r1) before decrementing it.
213 * The exec code can write slightly over 640kB to the stack
214 * before setting the user r1. Thus we allow the stack to
215 * expand to 1MB without further checks.
217 if (address + 0x100000 < vma->vm_end) {
218 /* get user regs even if this fault is in kernel mode */
219 struct pt_regs *uregs = current->thread.regs;
224 * A user-mode access to an address a long way below
225 * the stack pointer is only valid if the instruction
226 * is one which would update the stack pointer to the
227 * address accessed if the instruction completed,
228 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
229 * (or the byte, halfword, float or double forms).
231 * If we don't check this then any write to the area
232 * between the last mapped region and the stack will
233 * expand the stack rather than segfaulting.
235 if (address + 2048 < uregs->gpr[1]
236 && (!user_mode(regs) || !store_updates_sp(regs)))
239 if (expand_stack(vma, address))
244 #if defined(CONFIG_6xx)
245 if (error_code & 0x95700000)
246 /* an error such as lwarx to I/O controller space,
247 address matching DABR, eciwx, etc. */
249 #endif /* CONFIG_6xx */
250 #if defined(CONFIG_8xx)
251 /* The MPC8xx seems to always set 0x80000000, which is
252 * "undefined". Of those that can be set, this is the only
253 * one which seems bad.
255 if (error_code & 0x10000000)
256 /* Guarded storage error. */
258 #endif /* CONFIG_8xx */
262 /* protection fault */
263 if (error_code & DSISR_PROTFAULT)
265 if (!(vma->vm_flags & VM_EXEC))
268 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
272 /* Since 4xx/Book-E supports per-page execute permission,
273 * we lazily flush dcache to icache. */
275 if (get_pteptr(mm, address, &ptep, &pmdp)) {
276 spinlock_t *ptl = pte_lockptr(mm, pmdp);
278 if (pte_present(*ptep)) {
279 struct page *page = pte_page(*ptep);
281 if (!test_bit(PG_arch_1, &page->flags)) {
282 flush_dcache_icache_page(page);
283 set_bit(PG_arch_1, &page->flags);
285 pte_update(ptep, 0, _PAGE_HWEXEC);
287 pte_unmap_unlock(ptep, ptl);
288 up_read(&mm->mmap_sem);
291 pte_unmap_unlock(ptep, ptl);
295 } else if (is_write) {
296 if (!(vma->vm_flags & VM_WRITE))
300 /* protection fault */
301 if (error_code & 0x08000000)
303 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
308 * If for any reason at all we couldn't handle the fault,
309 * make sure we exit gracefully rather than endlessly redo
313 switch (handle_mm_fault(mm, vma, address, is_write)) {
321 case VM_FAULT_SIGBUS:
329 up_read(&mm->mmap_sem);
333 up_read(&mm->mmap_sem);
335 bad_area_nosemaphore:
336 /* User mode accesses cause a SIGSEGV */
337 if (user_mode(regs)) {
338 _exception(SIGSEGV, regs, code, address);
342 if (is_exec && (error_code & DSISR_PROTFAULT)
343 && printk_ratelimit())
344 printk(KERN_CRIT "kernel tried to execute NX-protected"
345 " page (%lx) - exploit attempt? (uid: %d)\n",
346 address, current->uid);
351 * We ran out of memory, or some other thing happened to us that made
352 * us unable to handle the page fault gracefully.
355 up_read(&mm->mmap_sem);
356 if (current->pid == 1) {
358 down_read(&mm->mmap_sem);
361 printk("VM: killing process %s\n", current->comm);
367 up_read(&mm->mmap_sem);
368 if (user_mode(regs)) {
369 info.si_signo = SIGBUS;
371 info.si_code = BUS_ADRERR;
372 info.si_addr = (void __user *)address;
373 force_sig_info(SIGBUS, &info, current);
380 * bad_page_fault is called when we have a bad access from the kernel.
381 * It is called from the DSI and ISI handlers in head.S and from some
382 * of the procedures in traps.c.
384 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
386 const struct exception_table_entry *entry;
388 /* Are we prepared to handle this fault? */
389 if ((entry = search_exception_tables(regs->nip)) != NULL) {
390 regs->nip = entry->fixup;
394 /* kernel has accessed a bad area */
396 printk(KERN_ALERT "Unable to handle kernel paging request for ");
397 switch (regs->trap) {
400 printk("data at address 0x%08lx\n", regs->dar);
404 printk("instruction fetch\n");
407 printk("unknown fault\n");
409 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
412 die("Kernel access of bad area", regs, sig);