]>
git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - arch/powerpc/mm/fault.c
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 error_code
)
88 if (notify_die(DIE_DABR_MATCH
, "dabr_match", regs
, error_code
,
89 11, SIGSEGV
) == NOTIFY_STOP
)
92 if (debugger_dabr_match(regs
))
98 /* Deliver the signal to userspace */
99 info
.si_signo
= SIGTRAP
;
101 info
.si_code
= TRAP_HWBKPT
;
102 info
.si_addr
= (void __user
*)regs
->nip
;
103 force_sig_info(SIGTRAP
, &info
, current
);
105 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
108 * For 600- and 800-family processors, the error_code parameter is DSISR
109 * for a data fault, SRR1 for an instruction fault. For 400-family processors
110 * the error_code parameter is ESR for a data fault, 0 for an instruction
112 * For 64-bit processors, the error_code parameter is
113 * - DSISR for a non-SLB data access fault,
114 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
117 * The return value is 0 if the fault was handled, or the signal
118 * number if this is a kernel fault that can't be handled here.
120 int __kprobes
do_page_fault(struct pt_regs
*regs
, unsigned long address
,
121 unsigned long error_code
)
123 struct vm_area_struct
* vma
;
124 struct mm_struct
*mm
= current
->mm
;
126 int code
= SEGV_MAPERR
;
128 int trap
= TRAP(regs
);
129 int is_exec
= trap
== 0x400;
131 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
133 * Fortunately the bit assignments in SRR1 for an instruction
134 * fault and DSISR for a data fault are mostly the same for the
135 * bits we are interested in. But there are some bits which
136 * indicate errors in DSISR but can validly be set in SRR1.
139 error_code
&= 0x48200000;
141 is_write
= error_code
& DSISR_ISSTORE
;
143 is_write
= error_code
& ESR_DST
;
144 #endif /* CONFIG_4xx || CONFIG_BOOKE */
146 if (notify_die(DIE_PAGE_FAULT
, "page_fault", regs
, error_code
,
147 11, SIGSEGV
) == NOTIFY_STOP
)
151 if (debugger_fault_handler(regs
))
155 /* On a kernel SLB miss we can only check for a valid exception entry */
156 if (!user_mode(regs
) && (address
>= TASK_SIZE
))
159 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
160 if (error_code
& DSISR_DABRMATCH
) {
162 do_dabr(regs
, error_code
);
165 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
167 if (in_atomic() || mm
== NULL
) {
168 if (!user_mode(regs
))
170 /* in_atomic() in user mode is really bad,
171 as is current->mm == NULL. */
172 printk(KERN_EMERG
"Page fault in user mode with"
173 "in_atomic() = %d mm = %p\n", in_atomic(), mm
);
174 printk(KERN_EMERG
"NIP = %lx MSR = %lx\n",
175 regs
->nip
, regs
->msr
);
176 die("Weird page fault", regs
, SIGSEGV
);
179 /* When running in the kernel we expect faults to occur only to
180 * addresses in user space. All other faults represent errors in the
181 * kernel and should generate an OOPS. Unfortunatly, in the case of an
182 * erroneous fault occuring in a code path which already holds mmap_sem
183 * we will deadlock attempting to validate the fault against the
184 * address space. Luckily the kernel only validly references user
185 * space from well defined areas of code, which are listed in the
188 * As the vast majority of faults will be valid we will only perform
189 * the source reference check when there is a possibilty of a deadlock.
190 * Attempt to lock the address space, if we cannot we then validate the
191 * source. If this is invalid we can skip the address space check,
192 * thus avoiding the deadlock.
194 if (!down_read_trylock(&mm
->mmap_sem
)) {
195 if (!user_mode(regs
) && !search_exception_tables(regs
->nip
))
196 goto bad_area_nosemaphore
;
198 down_read(&mm
->mmap_sem
);
201 vma
= find_vma(mm
, address
);
204 if (vma
->vm_start
<= address
)
206 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
210 * N.B. The POWER/Open ABI allows programs to access up to
211 * 288 bytes below the stack pointer.
212 * The kernel signal delivery code writes up to about 1.5kB
213 * below the stack pointer (r1) before decrementing it.
214 * The exec code can write slightly over 640kB to the stack
215 * before setting the user r1. Thus we allow the stack to
216 * expand to 1MB without further checks.
218 if (address
+ 0x100000 < vma
->vm_end
) {
219 /* get user regs even if this fault is in kernel mode */
220 struct pt_regs
*uregs
= current
->thread
.regs
;
225 * A user-mode access to an address a long way below
226 * the stack pointer is only valid if the instruction
227 * is one which would update the stack pointer to the
228 * address accessed if the instruction completed,
229 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
230 * (or the byte, halfword, float or double forms).
232 * If we don't check this then any write to the area
233 * between the last mapped region and the stack will
234 * expand the stack rather than segfaulting.
236 if (address
+ 2048 < uregs
->gpr
[1]
237 && (!user_mode(regs
) || !store_updates_sp(regs
)))
240 if (expand_stack(vma
, address
))
245 #if defined(CONFIG_6xx)
246 if (error_code
& 0x95700000)
247 /* an error such as lwarx to I/O controller space,
248 address matching DABR, eciwx, etc. */
250 #endif /* CONFIG_6xx */
251 #if defined(CONFIG_8xx)
252 /* The MPC8xx seems to always set 0x80000000, which is
253 * "undefined". Of those that can be set, this is the only
254 * one which seems bad.
256 if (error_code
& 0x10000000)
257 /* Guarded storage error. */
259 #endif /* CONFIG_8xx */
263 /* protection fault */
264 if (error_code
& DSISR_PROTFAULT
)
266 if (!(vma
->vm_flags
& VM_EXEC
))
269 #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
) && pte_present(*ptep
)) {
276 struct page
*page
= pte_page(*ptep
);
278 if (! test_bit(PG_arch_1
, &page
->flags
)) {
279 flush_dcache_icache_page(page
);
280 set_bit(PG_arch_1
, &page
->flags
);
282 pte_update(ptep
, 0, _PAGE_HWEXEC
);
285 up_read(&mm
->mmap_sem
);
292 } else if (is_write
) {
293 if (!(vma
->vm_flags
& VM_WRITE
))
297 /* protection fault */
298 if (error_code
& 0x08000000)
300 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
)))
305 * If for any reason at all we couldn't handle the fault,
306 * make sure we exit gracefully rather than endlessly redo
310 switch (handle_mm_fault(mm
, vma
, address
, is_write
)) {
318 case VM_FAULT_SIGBUS
:
326 up_read(&mm
->mmap_sem
);
330 up_read(&mm
->mmap_sem
);
332 bad_area_nosemaphore
:
333 /* User mode accesses cause a SIGSEGV */
334 if (user_mode(regs
)) {
335 _exception(SIGSEGV
, regs
, code
, address
);
339 if (is_exec
&& (error_code
& DSISR_PROTFAULT
)
340 && printk_ratelimit())
341 printk(KERN_CRIT
"kernel tried to execute NX-protected"
342 " page (%lx) - exploit attempt? (uid: %d)\n",
343 address
, current
->uid
);
348 * We ran out of memory, or some other thing happened to us that made
349 * us unable to handle the page fault gracefully.
352 up_read(&mm
->mmap_sem
);
353 if (current
->pid
== 1) {
355 down_read(&mm
->mmap_sem
);
358 printk("VM: killing process %s\n", current
->comm
);
364 up_read(&mm
->mmap_sem
);
365 if (user_mode(regs
)) {
366 info
.si_signo
= SIGBUS
;
368 info
.si_code
= BUS_ADRERR
;
369 info
.si_addr
= (void __user
*)address
;
370 force_sig_info(SIGBUS
, &info
, current
);
377 * bad_page_fault is called when we have a bad access from the kernel.
378 * It is called from the DSI and ISI handlers in head.S and from some
379 * of the procedures in traps.c.
381 void bad_page_fault(struct pt_regs
*regs
, unsigned long address
, int sig
)
383 const struct exception_table_entry
*entry
;
385 /* Are we prepared to handle this fault? */
386 if ((entry
= search_exception_tables(regs
->nip
)) != NULL
) {
387 regs
->nip
= entry
->fixup
;
391 /* kernel has accessed a bad area */
393 printk(KERN_ALERT
"Unable to handle kernel paging request for ");
394 switch (regs
->trap
) {
397 printk("data at address 0x%08lx\n", regs
->dar
);
401 printk("instruction fetch\n");
404 printk("unknown fault\n");
406 printk(KERN_ALERT
"Faulting instruction address: 0x%08lx\n",
409 die("Kernel access of bad area", regs
, sig
);