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>
32 #include <linux/perf_counter.h>
34 #include <asm/firmware.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/siginfo.h>
46 static inline int notify_page_fault(struct pt_regs
*regs
)
50 /* kprobe_running() needs smp_processor_id() */
51 if (!user_mode(regs
)) {
53 if (kprobe_running() && kprobe_fault_handler(regs
, 11))
61 static inline int notify_page_fault(struct pt_regs
*regs
)
68 * Check whether the instruction at regs->nip is a store using
69 * an update addressing form which will update r1.
71 static int store_updates_sp(struct pt_regs
*regs
)
75 if (get_user(inst
, (unsigned int __user
*)regs
->nip
))
77 /* check for 1 in the rA field */
78 if (((inst
>> 16) & 0x1f) != 1)
80 /* check major opcode */
88 case 62: /* std or stdu */
89 return (inst
& 3) == 1;
91 /* check minor opcode */
92 switch ((inst
>> 1) & 0x3ff) {
97 case 695: /* stfsux */
98 case 759: /* stfdux */
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
;
125 int is_write
= 0, ret
;
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_page_fault(regs
))
147 if (unlikely(debugger_fault_handler(regs
)))
150 /* On a kernel SLB miss we can only check for a valid exception entry */
151 if (!user_mode(regs
) && (address
>= TASK_SIZE
))
154 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
155 if (error_code
& DSISR_DABRMATCH
) {
157 do_dabr(regs
, address
, error_code
);
160 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
162 if (in_atomic() || mm
== NULL
|| current
->pagefault_disabled
) {
163 if (!user_mode(regs
))
165 /* in_atomic() in user mode is really bad,
166 as is current->mm == NULL. */
167 printk(KERN_EMERG
"Page fault in user mode with "
168 "in_atomic() = %d mm = %p\n", in_atomic(), mm
);
169 printk(KERN_EMERG
"NIP = %lx MSR = %lx\n",
170 regs
->nip
, regs
->msr
);
171 die("Weird page fault", regs
, SIGSEGV
);
174 perf_swcounter_event(PERF_COUNT_PAGE_FAULTS
, 1, 0, regs
);
176 /* When running in the kernel we expect faults to occur only to
177 * addresses in user space. All other faults represent errors in the
178 * kernel and should generate an OOPS. Unfortunately, in the case of an
179 * erroneous fault occurring in a code path which already holds mmap_sem
180 * we will deadlock attempting to validate the fault against the
181 * address space. Luckily the kernel only validly references user
182 * space from well defined areas of code, which are listed in the
185 * As the vast majority of faults will be valid we will only perform
186 * the source reference check when there is a possibility of a deadlock.
187 * Attempt to lock the address space, if we cannot we then validate the
188 * source. If this is invalid we can skip the address space check,
189 * thus avoiding the deadlock.
191 if (!down_read_trylock(&mm
->mmap_sem
)) {
192 if (!user_mode(regs
) && !search_exception_tables(regs
->nip
))
193 goto bad_area_nosemaphore
;
195 down_read(&mm
->mmap_sem
);
198 vma
= find_vma(mm
, address
);
201 if (vma
->vm_start
<= address
)
203 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
207 * N.B. The POWER/Open ABI allows programs to access up to
208 * 288 bytes below the stack pointer.
209 * The kernel signal delivery code writes up to about 1.5kB
210 * below the stack pointer (r1) before decrementing it.
211 * The exec code can write slightly over 640kB to the stack
212 * before setting the user r1. Thus we allow the stack to
213 * expand to 1MB without further checks.
215 if (address
+ 0x100000 < vma
->vm_end
) {
216 /* get user regs even if this fault is in kernel mode */
217 struct pt_regs
*uregs
= current
->thread
.regs
;
222 * A user-mode access to an address a long way below
223 * the stack pointer is only valid if the instruction
224 * is one which would update the stack pointer to the
225 * address accessed if the instruction completed,
226 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
227 * (or the byte, halfword, float or double forms).
229 * If we don't check this then any write to the area
230 * between the last mapped region and the stack will
231 * expand the stack rather than segfaulting.
233 if (address
+ 2048 < uregs
->gpr
[1]
234 && (!user_mode(regs
) || !store_updates_sp(regs
)))
237 if (expand_stack(vma
, address
))
242 #if defined(CONFIG_6xx)
243 if (error_code
& 0x95700000)
244 /* an error such as lwarx to I/O controller space,
245 address matching DABR, eciwx, etc. */
247 #endif /* CONFIG_6xx */
248 #if defined(CONFIG_8xx)
249 /* The MPC8xx seems to always set 0x80000000, which is
250 * "undefined". Of those that can be set, this is the only
251 * one which seems bad.
253 if (error_code
& 0x10000000)
254 /* Guarded storage error. */
256 #endif /* CONFIG_8xx */
259 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
260 /* protection fault */
261 if (error_code
& DSISR_PROTFAULT
)
264 * Allow execution from readable areas if the MMU does not
265 * provide separate controls over reading and executing.
267 if (!(vma
->vm_flags
& VM_EXEC
) &&
268 (cpu_has_feature(CPU_FTR_NOEXECUTE
) ||
269 !(vma
->vm_flags
& (VM_READ
| VM_WRITE
))))
275 /* Since 4xx/Book-E supports per-page execute permission,
276 * we lazily flush dcache to icache. */
278 if (get_pteptr(mm
, address
, &ptep
, &pmdp
)) {
279 spinlock_t
*ptl
= pte_lockptr(mm
, pmdp
);
281 if (pte_present(*ptep
)) {
282 struct page
*page
= pte_page(*ptep
);
284 if (!test_bit(PG_arch_1
, &page
->flags
)) {
285 flush_dcache_icache_page(page
);
286 set_bit(PG_arch_1
, &page
->flags
);
288 pte_update(ptep
, 0, _PAGE_HWEXEC
|
290 local_flush_tlb_page(vma
, address
);
291 pte_unmap_unlock(ptep
, ptl
);
292 up_read(&mm
->mmap_sem
);
295 pte_unmap_unlock(ptep
, ptl
);
299 } else if (is_write
) {
300 if (!(vma
->vm_flags
& VM_WRITE
))
304 /* protection fault */
305 if (error_code
& 0x08000000)
307 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
)))
312 * If for any reason at all we couldn't handle the fault,
313 * make sure we exit gracefully rather than endlessly redo
317 ret
= handle_mm_fault(mm
, vma
, address
, is_write
);
318 if (unlikely(ret
& VM_FAULT_ERROR
)) {
319 if (ret
& VM_FAULT_OOM
)
321 else if (ret
& VM_FAULT_SIGBUS
)
325 if (ret
& VM_FAULT_MAJOR
) {
327 perf_swcounter_event(PERF_COUNT_PAGE_FAULTS_MAJ
, 1, 0, regs
);
328 #ifdef CONFIG_PPC_SMLPAR
329 if (firmware_has_feature(FW_FEATURE_CMO
)) {
331 get_lppaca()->page_ins
+= (1 << PAGE_FACTOR
);
337 perf_swcounter_event(PERF_COUNT_PAGE_FAULTS_MIN
, 1, 0, regs
);
339 up_read(&mm
->mmap_sem
);
343 up_read(&mm
->mmap_sem
);
345 bad_area_nosemaphore
:
346 /* User mode accesses cause a SIGSEGV */
347 if (user_mode(regs
)) {
348 _exception(SIGSEGV
, regs
, code
, address
);
352 if (is_exec
&& (error_code
& DSISR_PROTFAULT
)
353 && printk_ratelimit())
354 printk(KERN_CRIT
"kernel tried to execute NX-protected"
355 " page (%lx) - exploit attempt? (uid: %d)\n",
356 address
, current_uid());
361 * We ran out of memory, or some other thing happened to us that made
362 * us unable to handle the page fault gracefully.
365 up_read(&mm
->mmap_sem
);
366 if (is_global_init(current
)) {
368 down_read(&mm
->mmap_sem
);
371 printk("VM: killing process %s\n", current
->comm
);
373 do_group_exit(SIGKILL
);
377 up_read(&mm
->mmap_sem
);
378 if (user_mode(regs
)) {
379 info
.si_signo
= SIGBUS
;
381 info
.si_code
= BUS_ADRERR
;
382 info
.si_addr
= (void __user
*)address
;
383 force_sig_info(SIGBUS
, &info
, current
);
390 * bad_page_fault is called when we have a bad access from the kernel.
391 * It is called from the DSI and ISI handlers in head.S and from some
392 * of the procedures in traps.c.
394 void bad_page_fault(struct pt_regs
*regs
, unsigned long address
, int sig
)
396 const struct exception_table_entry
*entry
;
398 /* Are we prepared to handle this fault? */
399 if ((entry
= search_exception_tables(regs
->nip
)) != NULL
) {
400 regs
->nip
= entry
->fixup
;
404 /* kernel has accessed a bad area */
406 switch (regs
->trap
) {
409 printk(KERN_ALERT
"Unable to handle kernel paging request for "
410 "data at address 0x%08lx\n", regs
->dar
);
414 printk(KERN_ALERT
"Unable to handle kernel paging request for "
415 "instruction fetch\n");
418 printk(KERN_ALERT
"Unable to handle kernel paging request for "
422 printk(KERN_ALERT
"Faulting instruction address: 0x%08lx\n",
425 die("Kernel access of bad area", regs
, sig
);