1 /*
2 * arch/microblaze/mm/fault.c
3 *
4 * Copyright (C) 2007 Xilinx, Inc. All rights reserved.
5 *
6 * Derived from "arch/ppc/mm/fault.c"
7 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
8 *
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
11 *
12 * Modified by Cort Dougan and Paul Mackerras.
13 *
14 * This file is subject to the terms and conditions of the GNU General
15 * Public License. See the file COPYING in the main directory of this
16 * archive for more details.
17 *
18 */
19
20 #include <linux/extable.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>
29 #include <linux/mm.h>
30 #include <linux/interrupt.h>
31
32 #include <asm/page.h>
33 #include <asm/pgtable.h>
34 #include <asm/mmu.h>
35 #include <linux/mmu_context.h>
36 #include <linux/uaccess.h>
37 #include <asm/exceptions.h>
38
39 static unsigned long pte_misses; /* updated by do_page_fault() */
40 static unsigned long pte_errors; /* updated by do_page_fault() */
41
42 /*
43 * Check whether the instruction at regs->pc is a store using
44 * an update addressing form which will update r1.
45 */
46 static int store_updates_sp(struct pt_regs *regs)
47 {
48 unsigned int inst;
49
50 if (get_user(inst, (unsigned int __user *)regs->pc))
51 return 0;
52 /* check for 1 in the rD field */
53 if (((inst >> 21) & 0x1f) != 1)
54 return 0;
55 /* check for store opcodes */
56 if ((inst & 0xd0000000) == 0xd0000000)
57 return 1;
58 return 0;
59 }
60
61
62 /*
63 * bad_page_fault is called when we have a bad access from the kernel.
64 * It is called from do_page_fault above and from some of the procedures
65 * in traps.c.
66 */
67 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
68 {
69 const struct exception_table_entry *fixup;
70 /* MS: no context */
71 /* Are we prepared to handle this fault? */
72 fixup = search_exception_tables(regs->pc);
73 if (fixup) {
74 regs->pc = fixup->fixup;
75 return;
76 }
77
78 /* kernel has accessed a bad area */
79 die("kernel access of bad area", regs, sig);
80 }
81
82 /*
83 * The error_code parameter is ESR for a data fault,
84 * 0 for an instruction fault.
85 */
86 void do_page_fault(struct pt_regs *regs, unsigned long address,
87 unsigned long error_code)
88 {
89 struct vm_area_struct *vma;
90 struct mm_struct *mm = current->mm;
91 int code = SEGV_MAPERR;
92 int is_write = error_code & ESR_S;
93 vm_fault_t fault;
94 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
95
96 regs->ear = address;
97 regs->esr = error_code;
98
99 /* On a kernel SLB miss we can only check for a valid exception entry */
100 if (unlikely(kernel_mode(regs) && (address >= TASK_SIZE))) {
101 pr_warn("kernel task_size exceed");
102 _exception(SIGSEGV, regs, code, address);
103 }
104
105 /* for instr TLB miss and instr storage exception ESR_S is undefined */
106 if ((error_code & 0x13) == 0x13 || (error_code & 0x11) == 0x11)
107 is_write = 0;
108
109 if (unlikely(faulthandler_disabled() || !mm)) {
110 if (kernel_mode(regs))
111 goto bad_area_nosemaphore;
112
113 /* faulthandler_disabled() in user mode is really bad,
114 as is current->mm == NULL. */
115 pr_emerg("Page fault in user mode with faulthandler_disabled(), mm = %p\n",
116 mm);
117 pr_emerg("r15 = %lx MSR = %lx\n",
118 regs->r15, regs->msr);
119 die("Weird page fault", regs, SIGSEGV);
120 }
121
122 if (user_mode(regs))
123 flags |= FAULT_FLAG_USER;
124
125 /* When running in the kernel we expect faults to occur only to
126 * addresses in user space. All other faults represent errors in the
127 * kernel and should generate an OOPS. Unfortunately, in the case of an
128 * erroneous fault occurring in a code path which already holds mmap_sem
129 * we will deadlock attempting to validate the fault against the
130 * address space. Luckily the kernel only validly references user
131 * space from well defined areas of code, which are listed in the
132 * exceptions table.
133 *
134 * As the vast majority of faults will be valid we will only perform
135 * the source reference check when there is a possibility of a deadlock.
136 * Attempt to lock the address space, if we cannot we then validate the
137 * source. If this is invalid we can skip the address space check,
138 * thus avoiding the deadlock.
139 */
140 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
141 if (kernel_mode(regs) && !search_exception_tables(regs->pc))
142 goto bad_area_nosemaphore;
143
144 retry:
145 down_read(&mm->mmap_sem);
146 }
147
148 vma = find_vma(mm, address);
149 if (unlikely(!vma))
150 goto bad_area;
151
152 if (vma->vm_start <= address)
153 goto good_area;
154
155 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
156 goto bad_area;
157
158 if (unlikely(!is_write))
159 goto bad_area;
160
161 /*
162 * N.B. The ABI allows programs to access up to
163 * a few hundred bytes below the stack pointer (TBD).
164 * The kernel signal delivery code writes up to about 1.5kB
165 * below the stack pointer (r1) before decrementing it.
166 * The exec code can write slightly over 640kB to the stack
167 * before setting the user r1. Thus we allow the stack to
168 * expand to 1MB without further checks.
169 */
170 if (unlikely(address + 0x100000 < vma->vm_end)) {
171
172 /* get user regs even if this fault is in kernel mode */
173 struct pt_regs *uregs = current->thread.regs;
174 if (uregs == NULL)
175 goto bad_area;
176
177 /*
178 * A user-mode access to an address a long way below
179 * the stack pointer is only valid if the instruction
180 * is one which would update the stack pointer to the
181 * address accessed if the instruction completed,
182 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
183 * (or the byte, halfword, float or double forms).
184 *
185 * If we don't check this then any write to the area
186 * between the last mapped region and the stack will
187 * expand the stack rather than segfaulting.
188 */
189 if (address + 2048 < uregs->r1
190 && (kernel_mode(regs) || !store_updates_sp(regs)))
191 goto bad_area;
192 }
193 if (expand_stack(vma, address))
194 goto bad_area;
195
196 good_area:
197 code = SEGV_ACCERR;
198
199 /* a write */
200 if (unlikely(is_write)) {
201 if (unlikely(!(vma->vm_flags & VM_WRITE)))
202 goto bad_area;
203 flags |= FAULT_FLAG_WRITE;
204 /* a read */
205 } else {
206 /* protection fault */
207 if (unlikely(error_code & 0x08000000))
208 goto bad_area;
209 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC))))
210 goto bad_area;
211 }
212
213 /*
214 * If for any reason at all we couldn't handle the fault,
215 * make sure we exit gracefully rather than endlessly redo
216 * the fault.
217 */
218 fault = handle_mm_fault(vma, address, flags);
219
220 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
221 return;
222
223 if (unlikely(fault & VM_FAULT_ERROR)) {
224 if (fault & VM_FAULT_OOM)
225 goto out_of_memory;
226 else if (fault & VM_FAULT_SIGSEGV)
227 goto bad_area;
228 else if (fault & VM_FAULT_SIGBUS)
229 goto do_sigbus;
230 BUG();
231 }
232
233 if (flags & FAULT_FLAG_ALLOW_RETRY) {
234 if (unlikely(fault & VM_FAULT_MAJOR))
235 current->maj_flt++;
236 else
237 current->min_flt++;
238 if (fault & VM_FAULT_RETRY) {
239 flags &= ~FAULT_FLAG_ALLOW_RETRY;
240 flags |= FAULT_FLAG_TRIED;
241
242 /*
243 * No need to up_read(&mm->mmap_sem) as we would
244 * have already released it in __lock_page_or_retry
245 * in mm/filemap.c.
246 */
247
248 goto retry;
249 }
250 }
251
252 up_read(&mm->mmap_sem);
253
254 /*
255 * keep track of tlb+htab misses that are good addrs but
256 * just need pte's created via handle_mm_fault()
257 * -- Cort
258 */
259 pte_misses++;
260 return;
261
262 bad_area:
263 up_read(&mm->mmap_sem);
264
265 bad_area_nosemaphore:
266 pte_errors++;
267
268 /* User mode accesses cause a SIGSEGV */
269 if (user_mode(regs)) {
270 _exception(SIGSEGV, regs, code, address);
271 return;
272 }
273
274 bad_page_fault(regs, address, SIGSEGV);
275 return;
276
277 /*
278 * We ran out of memory, or some other thing happened to us that made
279 * us unable to handle the page fault gracefully.
280 */
281 out_of_memory:
282 up_read(&mm->mmap_sem);
283 if (!user_mode(regs))
284 bad_page_fault(regs, address, SIGKILL);
285 else
286 pagefault_out_of_memory();
287 return;
288
289 do_sigbus:
290 up_read(&mm->mmap_sem);
291 if (user_mode(regs)) {
292 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
293 return;
294 }
295 bad_page_fault(regs, address, SIGBUS);
296 }