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/signal.h>
19 #include <linux/sched.h>
20 #include <linux/kernel.h>
21 #include <linux/errno.h>
22 #include <linux/string.h>
23 #include <linux/types.h>
24 #include <linux/ptrace.h>
25 #include <linux/mman.h>
27 #include <linux/interrupt.h>
28 #include <linux/highmem.h>
29 #include <linux/module.h>
30 #include <linux/kprobes.h>
31 #include <linux/kdebug.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/siginfo.h>
44 static inline int notify_page_fault(struct pt_regs *regs)
48 /* kprobe_running() needs smp_processor_id() */
49 if (!user_mode(regs)) {
51 if (kprobe_running() && kprobe_fault_handler(regs, 11))
59 static inline int notify_page_fault(struct pt_regs *regs)
66 * Check whether the instruction at regs->nip is a store using
67 * an update addressing form which will update r1.
69 static int store_updates_sp(struct pt_regs *regs)
73 if (get_user(inst, (unsigned int __user *)regs->nip))
75 /* check for 1 in the rA field */
76 if (((inst >> 16) & 0x1f) != 1)
78 /* check major opcode */
86 case 62: /* std or stdu */
87 return (inst & 3) == 1;
89 /* check minor opcode */
90 switch ((inst >> 1) & 0x3ff) {
95 case 695: /* stfsux */
96 case 759: /* stfdux */
103 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
104 static void do_dabr(struct pt_regs *regs, unsigned long address,
105 unsigned long error_code)
109 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
110 11, SIGSEGV) == NOTIFY_STOP)
113 if (debugger_dabr_match(regs))
119 /* Deliver the signal to userspace */
120 info.si_signo = SIGTRAP;
122 info.si_code = TRAP_HWBKPT;
123 info.si_addr = (void __user *)address;
124 force_sig_info(SIGTRAP, &info, current);
126 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
129 * For 600- and 800-family processors, the error_code parameter is DSISR
130 * for a data fault, SRR1 for an instruction fault. For 400-family processors
131 * the error_code parameter is ESR for a data fault, 0 for an instruction
133 * For 64-bit processors, the error_code parameter is
134 * - DSISR for a non-SLB data access fault,
135 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
138 * The return value is 0 if the fault was handled, or the signal
139 * number if this is a kernel fault that can't be handled here.
141 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
142 unsigned long error_code)
144 struct vm_area_struct * vma;
145 struct mm_struct *mm = current->mm;
147 int code = SEGV_MAPERR;
148 int is_write = 0, ret;
149 int trap = TRAP(regs);
150 int is_exec = trap == 0x400;
152 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
154 * Fortunately the bit assignments in SRR1 for an instruction
155 * fault and DSISR for a data fault are mostly the same for the
156 * bits we are interested in. But there are some bits which
157 * indicate errors in DSISR but can validly be set in SRR1.
160 error_code &= 0x48200000;
162 is_write = error_code & DSISR_ISSTORE;
164 is_write = error_code & ESR_DST;
165 #endif /* CONFIG_4xx || CONFIG_BOOKE */
167 if (notify_page_fault(regs))
171 if (debugger_fault_handler(regs))
175 /* On a kernel SLB miss we can only check for a valid exception entry */
176 if (!user_mode(regs) && (address >= TASK_SIZE))
179 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
180 if (error_code & DSISR_DABRMATCH) {
182 do_dabr(regs, address, error_code);
185 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
187 if (in_atomic() || mm == NULL) {
188 if (!user_mode(regs))
190 /* in_atomic() in user mode is really bad,
191 as is current->mm == NULL. */
192 printk(KERN_EMERG "Page fault in user mode with"
193 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
194 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
195 regs->nip, regs->msr);
196 die("Weird page fault", regs, SIGSEGV);
199 /* When running in the kernel we expect faults to occur only to
200 * addresses in user space. All other faults represent errors in the
201 * kernel and should generate an OOPS. Unfortunately, in the case of an
202 * erroneous fault occurring in a code path which already holds mmap_sem
203 * we will deadlock attempting to validate the fault against the
204 * address space. Luckily the kernel only validly references user
205 * space from well defined areas of code, which are listed in the
208 * As the vast majority of faults will be valid we will only perform
209 * the source reference check when there is a possibility of a deadlock.
210 * Attempt to lock the address space, if we cannot we then validate the
211 * source. If this is invalid we can skip the address space check,
212 * thus avoiding the deadlock.
214 if (!down_read_trylock(&mm->mmap_sem)) {
215 if (!user_mode(regs) && !search_exception_tables(regs->nip))
216 goto bad_area_nosemaphore;
218 down_read(&mm->mmap_sem);
221 vma = find_vma(mm, address);
224 if (vma->vm_start <= address)
226 if (!(vma->vm_flags & VM_GROWSDOWN))
230 * N.B. The POWER/Open ABI allows programs to access up to
231 * 288 bytes below the stack pointer.
232 * The kernel signal delivery code writes up to about 1.5kB
233 * below the stack pointer (r1) before decrementing it.
234 * The exec code can write slightly over 640kB to the stack
235 * before setting the user r1. Thus we allow the stack to
236 * expand to 1MB without further checks.
238 if (address + 0x100000 < vma->vm_end) {
239 /* get user regs even if this fault is in kernel mode */
240 struct pt_regs *uregs = current->thread.regs;
245 * A user-mode access to an address a long way below
246 * the stack pointer is only valid if the instruction
247 * is one which would update the stack pointer to the
248 * address accessed if the instruction completed,
249 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
250 * (or the byte, halfword, float or double forms).
252 * If we don't check this then any write to the area
253 * between the last mapped region and the stack will
254 * expand the stack rather than segfaulting.
256 if (address + 2048 < uregs->gpr[1]
257 && (!user_mode(regs) || !store_updates_sp(regs)))
260 if (expand_stack(vma, address))
265 #if defined(CONFIG_6xx)
266 if (error_code & 0x95700000)
267 /* an error such as lwarx to I/O controller space,
268 address matching DABR, eciwx, etc. */
270 #endif /* CONFIG_6xx */
271 #if defined(CONFIG_8xx)
272 /* The MPC8xx seems to always set 0x80000000, which is
273 * "undefined". Of those that can be set, this is the only
274 * one which seems bad.
276 if (error_code & 0x10000000)
277 /* Guarded storage error. */
279 #endif /* CONFIG_8xx */
282 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
283 /* protection fault */
284 if (error_code & DSISR_PROTFAULT)
286 if (!(vma->vm_flags & VM_EXEC))
292 /* Since 4xx/Book-E supports per-page execute permission,
293 * we lazily flush dcache to icache. */
295 if (get_pteptr(mm, address, &ptep, &pmdp)) {
296 spinlock_t *ptl = pte_lockptr(mm, pmdp);
298 if (pte_present(*ptep)) {
299 struct page *page = pte_page(*ptep);
301 if (!test_bit(PG_arch_1, &page->flags)) {
302 flush_dcache_icache_page(page);
303 set_bit(PG_arch_1, &page->flags);
305 pte_update(ptep, 0, _PAGE_HWEXEC);
307 pte_unmap_unlock(ptep, ptl);
308 up_read(&mm->mmap_sem);
311 pte_unmap_unlock(ptep, ptl);
315 } else if (is_write) {
316 if (!(vma->vm_flags & VM_WRITE))
320 /* protection fault */
321 if (error_code & 0x08000000)
323 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
328 * If for any reason at all we couldn't handle the fault,
329 * make sure we exit gracefully rather than endlessly redo
333 ret = handle_mm_fault(mm, vma, address, is_write);
334 if (unlikely(ret & VM_FAULT_ERROR)) {
335 if (ret & VM_FAULT_OOM)
337 else if (ret & VM_FAULT_SIGBUS)
341 if (ret & VM_FAULT_MAJOR)
345 up_read(&mm->mmap_sem);
349 up_read(&mm->mmap_sem);
351 bad_area_nosemaphore:
352 /* User mode accesses cause a SIGSEGV */
353 if (user_mode(regs)) {
354 _exception(SIGSEGV, regs, code, address);
358 if (is_exec && (error_code & DSISR_PROTFAULT)
359 && printk_ratelimit())
360 printk(KERN_CRIT "kernel tried to execute NX-protected"
361 " page (%lx) - exploit attempt? (uid: %d)\n",
362 address, current->uid);
367 * We ran out of memory, or some other thing happened to us that made
368 * us unable to handle the page fault gracefully.
371 up_read(&mm->mmap_sem);
372 if (is_init(current)) {
374 down_read(&mm->mmap_sem);
377 printk("VM: killing process %s\n", current->comm);
379 do_group_exit(SIGKILL);
383 up_read(&mm->mmap_sem);
384 if (user_mode(regs)) {
385 info.si_signo = SIGBUS;
387 info.si_code = BUS_ADRERR;
388 info.si_addr = (void __user *)address;
389 force_sig_info(SIGBUS, &info, current);
396 * bad_page_fault is called when we have a bad access from the kernel.
397 * It is called from the DSI and ISI handlers in head.S and from some
398 * of the procedures in traps.c.
400 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
402 const struct exception_table_entry *entry;
404 /* Are we prepared to handle this fault? */
405 if ((entry = search_exception_tables(regs->nip)) != NULL) {
406 regs->nip = entry->fixup;
410 /* kernel has accessed a bad area */
412 switch (regs->trap) {
415 printk(KERN_ALERT "Unable to handle kernel paging request for "
416 "data at address 0x%08lx\n", regs->dar);
420 printk(KERN_ALERT "Unable to handle kernel paging request for "
421 "instruction fetch\n");
424 printk(KERN_ALERT "Unable to handle kernel paging request for "
428 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
431 die("Kernel access of bad area", regs, sig);