Commit | Line | Data |
---|---|---|
14cf11af | 1 | /* |
14cf11af PM |
2 | * PowerPC version |
3 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) | |
4 | * | |
5 | * Derived from "arch/i386/mm/fault.c" | |
6 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
7 | * | |
8 | * Modified by Cort Dougan and Paul Mackerras. | |
9 | * | |
10 | * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com) | |
11 | * | |
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. | |
16 | */ | |
17 | ||
14cf11af PM |
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> | |
26 | #include <linux/mm.h> | |
27 | #include <linux/interrupt.h> | |
28 | #include <linux/highmem.h> | |
29 | #include <linux/module.h> | |
30 | #include <linux/kprobes.h> | |
1eeb66a1 | 31 | #include <linux/kdebug.h> |
cdd6c482 | 32 | #include <linux/perf_event.h> |
28b54990 | 33 | #include <linux/magic.h> |
14cf11af | 34 | |
40900194 | 35 | #include <asm/firmware.h> |
14cf11af PM |
36 | #include <asm/page.h> |
37 | #include <asm/pgtable.h> | |
38 | #include <asm/mmu.h> | |
39 | #include <asm/mmu_context.h> | |
40 | #include <asm/system.h> | |
41 | #include <asm/uaccess.h> | |
42 | #include <asm/tlbflush.h> | |
14cf11af | 43 | #include <asm/siginfo.h> |
5efab4a0 | 44 | #include <mm/mmu_decl.h> |
4f9e87c0 | 45 | |
9f90b997 CH |
46 | #ifdef CONFIG_KPROBES |
47 | static inline int notify_page_fault(struct pt_regs *regs) | |
4f9e87c0 | 48 | { |
9f90b997 CH |
49 | int ret = 0; |
50 | ||
51 | /* kprobe_running() needs smp_processor_id() */ | |
52 | if (!user_mode(regs)) { | |
53 | preempt_disable(); | |
54 | if (kprobe_running() && kprobe_fault_handler(regs, 11)) | |
55 | ret = 1; | |
56 | preempt_enable(); | |
57 | } | |
4f9e87c0 | 58 | |
9f90b997 | 59 | return ret; |
4f9e87c0 AK |
60 | } |
61 | #else | |
9f90b997 | 62 | static inline int notify_page_fault(struct pt_regs *regs) |
4f9e87c0 | 63 | { |
9f90b997 | 64 | return 0; |
4f9e87c0 AK |
65 | } |
66 | #endif | |
67 | ||
14cf11af PM |
68 | /* |
69 | * Check whether the instruction at regs->nip is a store using | |
70 | * an update addressing form which will update r1. | |
71 | */ | |
72 | static int store_updates_sp(struct pt_regs *regs) | |
73 | { | |
74 | unsigned int inst; | |
75 | ||
76 | if (get_user(inst, (unsigned int __user *)regs->nip)) | |
77 | return 0; | |
78 | /* check for 1 in the rA field */ | |
79 | if (((inst >> 16) & 0x1f) != 1) | |
80 | return 0; | |
81 | /* check major opcode */ | |
82 | switch (inst >> 26) { | |
83 | case 37: /* stwu */ | |
84 | case 39: /* stbu */ | |
85 | case 45: /* sthu */ | |
86 | case 53: /* stfsu */ | |
87 | case 55: /* stfdu */ | |
88 | return 1; | |
89 | case 62: /* std or stdu */ | |
90 | return (inst & 3) == 1; | |
91 | case 31: | |
92 | /* check minor opcode */ | |
93 | switch ((inst >> 1) & 0x3ff) { | |
94 | case 181: /* stdux */ | |
95 | case 183: /* stwux */ | |
96 | case 247: /* stbux */ | |
97 | case 439: /* sthux */ | |
98 | case 695: /* stfsux */ | |
99 | case 759: /* stfdux */ | |
100 | return 1; | |
101 | } | |
102 | } | |
103 | return 0; | |
104 | } | |
105 | ||
14cf11af PM |
106 | /* |
107 | * For 600- and 800-family processors, the error_code parameter is DSISR | |
108 | * for a data fault, SRR1 for an instruction fault. For 400-family processors | |
109 | * the error_code parameter is ESR for a data fault, 0 for an instruction | |
110 | * fault. | |
111 | * For 64-bit processors, the error_code parameter is | |
112 | * - DSISR for a non-SLB data access fault, | |
113 | * - SRR1 & 0x08000000 for a non-SLB instruction access fault | |
114 | * - 0 any SLB fault. | |
115 | * | |
116 | * The return value is 0 if the fault was handled, or the signal | |
117 | * number if this is a kernel fault that can't be handled here. | |
118 | */ | |
119 | int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address, | |
120 | unsigned long error_code) | |
121 | { | |
122 | struct vm_area_struct * vma; | |
123 | struct mm_struct *mm = current->mm; | |
124 | siginfo_t info; | |
125 | int code = SEGV_MAPERR; | |
83c54070 | 126 | int is_write = 0, ret; |
14cf11af PM |
127 | int trap = TRAP(regs); |
128 | int is_exec = trap == 0x400; | |
129 | ||
130 | #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) | |
131 | /* | |
132 | * Fortunately the bit assignments in SRR1 for an instruction | |
133 | * fault and DSISR for a data fault are mostly the same for the | |
134 | * bits we are interested in. But there are some bits which | |
135 | * indicate errors in DSISR but can validly be set in SRR1. | |
136 | */ | |
137 | if (trap == 0x400) | |
138 | error_code &= 0x48200000; | |
139 | else | |
140 | is_write = error_code & DSISR_ISSTORE; | |
141 | #else | |
142 | is_write = error_code & ESR_DST; | |
143 | #endif /* CONFIG_4xx || CONFIG_BOOKE */ | |
144 | ||
9f90b997 | 145 | if (notify_page_fault(regs)) |
14cf11af PM |
146 | return 0; |
147 | ||
c3b75bd7 MN |
148 | if (unlikely(debugger_fault_handler(regs))) |
149 | return 0; | |
14cf11af PM |
150 | |
151 | /* On a kernel SLB miss we can only check for a valid exception entry */ | |
152 | if (!user_mode(regs) && (address >= TASK_SIZE)) | |
153 | return SIGSEGV; | |
154 | ||
9c7cc234 P |
155 | #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \ |
156 | defined(CONFIG_PPC_BOOK3S_64)) | |
14cf11af PM |
157 | if (error_code & DSISR_DABRMATCH) { |
158 | /* DABR match */ | |
bce6c5fd | 159 | do_dabr(regs, address, error_code); |
14cf11af PM |
160 | return 0; |
161 | } | |
9c7cc234 | 162 | #endif |
14cf11af PM |
163 | |
164 | if (in_atomic() || mm == NULL) { | |
165 | if (!user_mode(regs)) | |
166 | return SIGSEGV; | |
167 | /* in_atomic() in user mode is really bad, | |
168 | as is current->mm == NULL. */ | |
df3c9019 | 169 | printk(KERN_EMERG "Page fault in user mode with " |
14cf11af PM |
170 | "in_atomic() = %d mm = %p\n", in_atomic(), mm); |
171 | printk(KERN_EMERG "NIP = %lx MSR = %lx\n", | |
172 | regs->nip, regs->msr); | |
173 | die("Weird page fault", regs, SIGSEGV); | |
174 | } | |
175 | ||
cdd6c482 | 176 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); |
7dd1fcc2 | 177 | |
14cf11af PM |
178 | /* When running in the kernel we expect faults to occur only to |
179 | * addresses in user space. All other faults represent errors in the | |
fc5266ea AB |
180 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
181 | * erroneous fault occurring in a code path which already holds mmap_sem | |
14cf11af PM |
182 | * we will deadlock attempting to validate the fault against the |
183 | * address space. Luckily the kernel only validly references user | |
184 | * space from well defined areas of code, which are listed in the | |
185 | * exceptions table. | |
186 | * | |
187 | * As the vast majority of faults will be valid we will only perform | |
fc5266ea | 188 | * the source reference check when there is a possibility of a deadlock. |
14cf11af PM |
189 | * Attempt to lock the address space, if we cannot we then validate the |
190 | * source. If this is invalid we can skip the address space check, | |
191 | * thus avoiding the deadlock. | |
192 | */ | |
193 | if (!down_read_trylock(&mm->mmap_sem)) { | |
194 | if (!user_mode(regs) && !search_exception_tables(regs->nip)) | |
195 | goto bad_area_nosemaphore; | |
196 | ||
197 | down_read(&mm->mmap_sem); | |
198 | } | |
199 | ||
200 | vma = find_vma(mm, address); | |
201 | if (!vma) | |
202 | goto bad_area; | |
203 | if (vma->vm_start <= address) | |
204 | goto good_area; | |
205 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
206 | goto bad_area; | |
207 | ||
208 | /* | |
209 | * N.B. The POWER/Open ABI allows programs to access up to | |
210 | * 288 bytes below the stack pointer. | |
211 | * The kernel signal delivery code writes up to about 1.5kB | |
212 | * below the stack pointer (r1) before decrementing it. | |
213 | * The exec code can write slightly over 640kB to the stack | |
214 | * before setting the user r1. Thus we allow the stack to | |
215 | * expand to 1MB without further checks. | |
216 | */ | |
217 | if (address + 0x100000 < vma->vm_end) { | |
218 | /* get user regs even if this fault is in kernel mode */ | |
219 | struct pt_regs *uregs = current->thread.regs; | |
220 | if (uregs == NULL) | |
221 | goto bad_area; | |
222 | ||
223 | /* | |
224 | * A user-mode access to an address a long way below | |
225 | * the stack pointer is only valid if the instruction | |
226 | * is one which would update the stack pointer to the | |
227 | * address accessed if the instruction completed, | |
228 | * i.e. either stwu rs,n(r1) or stwux rs,r1,rb | |
229 | * (or the byte, halfword, float or double forms). | |
230 | * | |
231 | * If we don't check this then any write to the area | |
232 | * between the last mapped region and the stack will | |
233 | * expand the stack rather than segfaulting. | |
234 | */ | |
235 | if (address + 2048 < uregs->gpr[1] | |
236 | && (!user_mode(regs) || !store_updates_sp(regs))) | |
237 | goto bad_area; | |
238 | } | |
239 | if (expand_stack(vma, address)) | |
240 | goto bad_area; | |
241 | ||
242 | good_area: | |
243 | code = SEGV_ACCERR; | |
244 | #if defined(CONFIG_6xx) | |
245 | if (error_code & 0x95700000) | |
246 | /* an error such as lwarx to I/O controller space, | |
247 | address matching DABR, eciwx, etc. */ | |
248 | goto bad_area; | |
249 | #endif /* CONFIG_6xx */ | |
250 | #if defined(CONFIG_8xx) | |
5efab4a0 JT |
251 | /* 8xx sometimes need to load a invalid/non-present TLBs. |
252 | * These must be invalidated separately as linux mm don't. | |
253 | */ | |
254 | if (error_code & 0x40000000) /* no translation? */ | |
255 | _tlbil_va(address, 0, 0, 0); | |
256 | ||
14cf11af PM |
257 | /* The MPC8xx seems to always set 0x80000000, which is |
258 | * "undefined". Of those that can be set, this is the only | |
259 | * one which seems bad. | |
260 | */ | |
261 | if (error_code & 0x10000000) | |
262 | /* Guarded storage error. */ | |
263 | goto bad_area; | |
264 | #endif /* CONFIG_8xx */ | |
265 | ||
266 | if (is_exec) { | |
8d30c14c BH |
267 | #ifdef CONFIG_PPC_STD_MMU |
268 | /* Protection fault on exec go straight to failure on | |
269 | * Hash based MMUs as they either don't support per-page | |
270 | * execute permission, or if they do, it's handled already | |
271 | * at the hash level. This test would probably have to | |
272 | * be removed if we change the way this works to make hash | |
273 | * processors use the same I/D cache coherency mechanism | |
274 | * as embedded. | |
275 | */ | |
14cf11af PM |
276 | if (error_code & DSISR_PROTFAULT) |
277 | goto bad_area; | |
8d30c14c BH |
278 | #endif /* CONFIG_PPC_STD_MMU */ |
279 | ||
08ae6cc1 PM |
280 | /* |
281 | * Allow execution from readable areas if the MMU does not | |
282 | * provide separate controls over reading and executing. | |
8d30c14c BH |
283 | * |
284 | * Note: That code used to not be enabled for 4xx/BookE. | |
285 | * It is now as I/D cache coherency for these is done at | |
286 | * set_pte_at() time and I see no reason why the test | |
287 | * below wouldn't be valid on those processors. This -may- | |
288 | * break programs compiled with a really old ABI though. | |
08ae6cc1 PM |
289 | */ |
290 | if (!(vma->vm_flags & VM_EXEC) && | |
291 | (cpu_has_feature(CPU_FTR_NOEXECUTE) || | |
292 | !(vma->vm_flags & (VM_READ | VM_WRITE)))) | |
14cf11af | 293 | goto bad_area; |
14cf11af PM |
294 | /* a write */ |
295 | } else if (is_write) { | |
296 | if (!(vma->vm_flags & VM_WRITE)) | |
297 | goto bad_area; | |
298 | /* a read */ | |
299 | } else { | |
300 | /* protection fault */ | |
301 | if (error_code & 0x08000000) | |
302 | goto bad_area; | |
df67b3da | 303 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) |
14cf11af PM |
304 | goto bad_area; |
305 | } | |
306 | ||
307 | /* | |
308 | * If for any reason at all we couldn't handle the fault, | |
309 | * make sure we exit gracefully rather than endlessly redo | |
310 | * the fault. | |
311 | */ | |
d06063cc | 312 | ret = handle_mm_fault(mm, vma, address, is_write ? FAULT_FLAG_WRITE : 0); |
83c54070 NP |
313 | if (unlikely(ret & VM_FAULT_ERROR)) { |
314 | if (ret & VM_FAULT_OOM) | |
315 | goto out_of_memory; | |
316 | else if (ret & VM_FAULT_SIGBUS) | |
317 | goto do_sigbus; | |
14cf11af PM |
318 | BUG(); |
319 | } | |
40900194 | 320 | if (ret & VM_FAULT_MAJOR) { |
83c54070 | 321 | current->maj_flt++; |
cdd6c482 | 322 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, |
78f13e95 | 323 | regs, address); |
40900194 BK |
324 | #ifdef CONFIG_PPC_SMLPAR |
325 | if (firmware_has_feature(FW_FEATURE_CMO)) { | |
326 | preempt_disable(); | |
a6326e98 | 327 | get_lppaca()->page_ins += (1 << PAGE_FACTOR); |
40900194 BK |
328 | preempt_enable(); |
329 | } | |
330 | #endif | |
ac17dc8e | 331 | } else { |
83c54070 | 332 | current->min_flt++; |
cdd6c482 | 333 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, |
78f13e95 | 334 | regs, address); |
ac17dc8e | 335 | } |
14cf11af PM |
336 | up_read(&mm->mmap_sem); |
337 | return 0; | |
338 | ||
339 | bad_area: | |
340 | up_read(&mm->mmap_sem); | |
341 | ||
342 | bad_area_nosemaphore: | |
343 | /* User mode accesses cause a SIGSEGV */ | |
344 | if (user_mode(regs)) { | |
345 | _exception(SIGSEGV, regs, code, address); | |
346 | return 0; | |
347 | } | |
348 | ||
349 | if (is_exec && (error_code & DSISR_PROTFAULT) | |
350 | && printk_ratelimit()) | |
351 | printk(KERN_CRIT "kernel tried to execute NX-protected" | |
352 | " page (%lx) - exploit attempt? (uid: %d)\n", | |
1330deb0 | 353 | address, current_uid()); |
14cf11af PM |
354 | |
355 | return SIGSEGV; | |
356 | ||
357 | /* | |
358 | * We ran out of memory, or some other thing happened to us that made | |
359 | * us unable to handle the page fault gracefully. | |
360 | */ | |
361 | out_of_memory: | |
362 | up_read(&mm->mmap_sem); | |
e460c2c9 BH |
363 | if (!user_mode(regs)) |
364 | return SIGKILL; | |
365 | pagefault_out_of_memory(); | |
366 | return 0; | |
14cf11af PM |
367 | |
368 | do_sigbus: | |
369 | up_read(&mm->mmap_sem); | |
370 | if (user_mode(regs)) { | |
371 | info.si_signo = SIGBUS; | |
372 | info.si_errno = 0; | |
373 | info.si_code = BUS_ADRERR; | |
374 | info.si_addr = (void __user *)address; | |
375 | force_sig_info(SIGBUS, &info, current); | |
376 | return 0; | |
377 | } | |
378 | return SIGBUS; | |
379 | } | |
380 | ||
381 | /* | |
382 | * bad_page_fault is called when we have a bad access from the kernel. | |
383 | * It is called from the DSI and ISI handlers in head.S and from some | |
384 | * of the procedures in traps.c. | |
385 | */ | |
386 | void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) | |
387 | { | |
388 | const struct exception_table_entry *entry; | |
28b54990 | 389 | unsigned long *stackend; |
14cf11af PM |
390 | |
391 | /* Are we prepared to handle this fault? */ | |
392 | if ((entry = search_exception_tables(regs->nip)) != NULL) { | |
393 | regs->nip = entry->fixup; | |
394 | return; | |
395 | } | |
396 | ||
397 | /* kernel has accessed a bad area */ | |
723925b7 | 398 | |
723925b7 | 399 | switch (regs->trap) { |
a416dd8d ME |
400 | case 0x300: |
401 | case 0x380: | |
402 | printk(KERN_ALERT "Unable to handle kernel paging request for " | |
403 | "data at address 0x%08lx\n", regs->dar); | |
404 | break; | |
405 | case 0x400: | |
406 | case 0x480: | |
407 | printk(KERN_ALERT "Unable to handle kernel paging request for " | |
408 | "instruction fetch\n"); | |
409 | break; | |
410 | default: | |
411 | printk(KERN_ALERT "Unable to handle kernel paging request for " | |
412 | "unknown fault\n"); | |
413 | break; | |
723925b7 OJ |
414 | } |
415 | printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", | |
416 | regs->nip); | |
417 | ||
28b54990 AB |
418 | stackend = end_of_stack(current); |
419 | if (current != &init_task && *stackend != STACK_END_MAGIC) | |
420 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); | |
421 | ||
14cf11af PM |
422 | die("Kernel access of bad area", regs, sig); |
423 | } |