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 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
84 static void do_dabr(struct pt_regs
*regs
, unsigned long address
,
85 unsigned long error_code
)
89 if (notify_die(DIE_DABR_MATCH
, "dabr_match", regs
, error_code
,
90 11, SIGSEGV
) == NOTIFY_STOP
)
93 if (debugger_dabr_match(regs
))
99 /* Deliver the signal to userspace */
100 info
.si_signo
= SIGTRAP
;
102 info
.si_code
= TRAP_HWBKPT
;
103 info
.si_addr
= (void __user
*)address
;
104 force_sig_info(SIGTRAP
, &info
, current
);
106 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
109 * For 600- and 800-family processors, the error_code parameter is DSISR
110 * for a data fault, SRR1 for an instruction fault. For 400-family processors
111 * the error_code parameter is ESR for a data fault, 0 for an instruction
113 * For 64-bit processors, the error_code parameter is
114 * - DSISR for a non-SLB data access fault,
115 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
118 * The return value is 0 if the fault was handled, or the signal
119 * number if this is a kernel fault that can't be handled here.
121 int __kprobes
do_page_fault(struct pt_regs
*regs
, unsigned long address
,
122 unsigned long error_code
)
124 struct vm_area_struct
* vma
;
125 struct mm_struct
*mm
= current
->mm
;
127 int code
= SEGV_MAPERR
;
129 int trap
= TRAP(regs
);
130 int is_exec
= trap
== 0x400;
132 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
134 * Fortunately the bit assignments in SRR1 for an instruction
135 * fault and DSISR for a data fault are mostly the same for the
136 * bits we are interested in. But there are some bits which
137 * indicate errors in DSISR but can validly be set in SRR1.
140 error_code
&= 0x48200000;
142 is_write
= error_code
& DSISR_ISSTORE
;
144 is_write
= error_code
& ESR_DST
;
145 #endif /* CONFIG_4xx || CONFIG_BOOKE */
147 if (notify_die(DIE_PAGE_FAULT
, "page_fault", regs
, error_code
,
148 11, SIGSEGV
) == NOTIFY_STOP
)
152 if (debugger_fault_handler(regs
))
156 /* On a kernel SLB miss we can only check for a valid exception entry */
157 if (!user_mode(regs
) && (address
>= TASK_SIZE
))
160 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
161 if (error_code
& DSISR_DABRMATCH
) {
163 do_dabr(regs
, address
, error_code
);
166 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
168 if (in_atomic() || mm
== NULL
) {
169 if (!user_mode(regs
))
171 /* in_atomic() in user mode is really bad,
172 as is current->mm == NULL. */
173 printk(KERN_EMERG
"Page fault in user mode with"
174 "in_atomic() = %d mm = %p\n", in_atomic(), mm
);
175 printk(KERN_EMERG
"NIP = %lx MSR = %lx\n",
176 regs
->nip
, regs
->msr
);
177 die("Weird page fault", regs
, SIGSEGV
);
180 /* When running in the kernel we expect faults to occur only to
181 * addresses in user space. All other faults represent errors in the
182 * kernel and should generate an OOPS. Unfortunatly, in the case of an
183 * erroneous fault occuring in a code path which already holds mmap_sem
184 * we will deadlock attempting to validate the fault against the
185 * address space. Luckily the kernel only validly references user
186 * space from well defined areas of code, which are listed in the
189 * As the vast majority of faults will be valid we will only perform
190 * the source reference check when there is a possibilty of a deadlock.
191 * Attempt to lock the address space, if we cannot we then validate the
192 * source. If this is invalid we can skip the address space check,
193 * thus avoiding the deadlock.
195 if (!down_read_trylock(&mm
->mmap_sem
)) {
196 if (!user_mode(regs
) && !search_exception_tables(regs
->nip
))
197 goto bad_area_nosemaphore
;
199 down_read(&mm
->mmap_sem
);
202 vma
= find_vma(mm
, address
);
205 if (vma
->vm_start
<= address
)
207 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
211 * N.B. The POWER/Open ABI allows programs to access up to
212 * 288 bytes below the stack pointer.
213 * The kernel signal delivery code writes up to about 1.5kB
214 * below the stack pointer (r1) before decrementing it.
215 * The exec code can write slightly over 640kB to the stack
216 * before setting the user r1. Thus we allow the stack to
217 * expand to 1MB without further checks.
219 if (address
+ 0x100000 < vma
->vm_end
) {
220 /* get user regs even if this fault is in kernel mode */
221 struct pt_regs
*uregs
= current
->thread
.regs
;
226 * A user-mode access to an address a long way below
227 * the stack pointer is only valid if the instruction
228 * is one which would update the stack pointer to the
229 * address accessed if the instruction completed,
230 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
231 * (or the byte, halfword, float or double forms).
233 * If we don't check this then any write to the area
234 * between the last mapped region and the stack will
235 * expand the stack rather than segfaulting.
237 if (address
+ 2048 < uregs
->gpr
[1]
238 && (!user_mode(regs
) || !store_updates_sp(regs
)))
241 if (expand_stack(vma
, address
))
246 #if defined(CONFIG_6xx)
247 if (error_code
& 0x95700000)
248 /* an error such as lwarx to I/O controller space,
249 address matching DABR, eciwx, etc. */
251 #endif /* CONFIG_6xx */
252 #if defined(CONFIG_8xx)
253 /* The MPC8xx seems to always set 0x80000000, which is
254 * "undefined". Of those that can be set, this is the only
255 * one which seems bad.
257 if (error_code
& 0x10000000)
258 /* Guarded storage error. */
260 #endif /* CONFIG_8xx */
264 /* protection fault */
265 if (error_code
& DSISR_PROTFAULT
)
267 if (!(vma
->vm_flags
& VM_EXEC
))
270 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
273 /* Since 4xx/Book-E supports per-page execute permission,
274 * we lazily flush dcache to icache. */
276 if (get_pteptr(mm
, address
, &ptep
) && pte_present(*ptep
)) {
277 struct page
*page
= pte_page(*ptep
);
279 if (! test_bit(PG_arch_1
, &page
->flags
)) {
280 flush_dcache_icache_page(page
);
281 set_bit(PG_arch_1
, &page
->flags
);
283 pte_update(ptep
, 0, _PAGE_HWEXEC
);
286 up_read(&mm
->mmap_sem
);
293 } else if (is_write
) {
294 if (!(vma
->vm_flags
& VM_WRITE
))
298 /* protection fault */
299 if (error_code
& 0x08000000)
301 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
)))
306 * If for any reason at all we couldn't handle the fault,
307 * make sure we exit gracefully rather than endlessly redo
311 switch (handle_mm_fault(mm
, vma
, address
, is_write
)) {
319 case VM_FAULT_SIGBUS
:
327 up_read(&mm
->mmap_sem
);
331 up_read(&mm
->mmap_sem
);
333 bad_area_nosemaphore
:
334 /* User mode accesses cause a SIGSEGV */
335 if (user_mode(regs
)) {
336 _exception(SIGSEGV
, regs
, code
, address
);
340 if (is_exec
&& (error_code
& DSISR_PROTFAULT
)
341 && printk_ratelimit())
342 printk(KERN_CRIT
"kernel tried to execute NX-protected"
343 " page (%lx) - exploit attempt? (uid: %d)\n",
344 address
, current
->uid
);
349 * We ran out of memory, or some other thing happened to us that made
350 * us unable to handle the page fault gracefully.
353 up_read(&mm
->mmap_sem
);
354 if (current
->pid
== 1) {
356 down_read(&mm
->mmap_sem
);
359 printk("VM: killing process %s\n", current
->comm
);
365 up_read(&mm
->mmap_sem
);
366 if (user_mode(regs
)) {
367 info
.si_signo
= SIGBUS
;
369 info
.si_code
= BUS_ADRERR
;
370 info
.si_addr
= (void __user
*)address
;
371 force_sig_info(SIGBUS
, &info
, current
);
378 * bad_page_fault is called when we have a bad access from the kernel.
379 * It is called from the DSI and ISI handlers in head.S and from some
380 * of the procedures in traps.c.
382 void bad_page_fault(struct pt_regs
*regs
, unsigned long address
, int sig
)
384 const struct exception_table_entry
*entry
;
386 /* Are we prepared to handle this fault? */
387 if ((entry
= search_exception_tables(regs
->nip
)) != NULL
) {
388 regs
->nip
= entry
->fixup
;
392 /* kernel has accessed a bad area */
394 printk(KERN_ALERT
"Unable to handle kernel paging request for ");
395 switch (regs
->trap
) {
398 printk("data at address 0x%08lx\n", regs
->dar
);
402 printk("instruction fetch\n");
405 printk("unknown fault\n");
407 printk(KERN_ALERT
"Faulting instruction address: 0x%08lx\n",
410 die("Kernel access of bad area", regs
, sig
);