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> | |
31 | ||
32 | #include <asm/page.h> | |
33 | #include <asm/pgtable.h> | |
34 | #include <asm/mmu.h> | |
35 | #include <asm/mmu_context.h> | |
36 | #include <asm/system.h> | |
37 | #include <asm/uaccess.h> | |
38 | #include <asm/tlbflush.h> | |
39 | #include <asm/kdebug.h> | |
40 | #include <asm/siginfo.h> | |
41 | ||
4f9e87c0 AK |
42 | #ifdef CONFIG_KPROBES |
43 | ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain); | |
44 | ||
45 | /* Hook to register for page fault notifications */ | |
46 | int register_page_fault_notifier(struct notifier_block *nb) | |
47 | { | |
48 | return atomic_notifier_chain_register(¬ify_page_fault_chain, nb); | |
49 | } | |
50 | ||
51 | int unregister_page_fault_notifier(struct notifier_block *nb) | |
52 | { | |
53 | return atomic_notifier_chain_unregister(¬ify_page_fault_chain, nb); | |
54 | } | |
55 | ||
56 | static inline int notify_page_fault(enum die_val val, const char *str, | |
57 | struct pt_regs *regs, long err, int trap, int sig) | |
58 | { | |
59 | struct die_args args = { | |
60 | .regs = regs, | |
61 | .str = str, | |
62 | .err = err, | |
63 | .trapnr = trap, | |
64 | .signr = sig | |
65 | }; | |
66 | return atomic_notifier_call_chain(¬ify_page_fault_chain, val, &args); | |
67 | } | |
68 | #else | |
69 | static inline int notify_page_fault(enum die_val val, const char *str, | |
70 | struct pt_regs *regs, long err, int trap, int sig) | |
71 | { | |
72 | return NOTIFY_DONE; | |
73 | } | |
74 | #endif | |
75 | ||
14cf11af PM |
76 | /* |
77 | * Check whether the instruction at regs->nip is a store using | |
78 | * an update addressing form which will update r1. | |
79 | */ | |
80 | static int store_updates_sp(struct pt_regs *regs) | |
81 | { | |
82 | unsigned int inst; | |
83 | ||
84 | if (get_user(inst, (unsigned int __user *)regs->nip)) | |
85 | return 0; | |
86 | /* check for 1 in the rA field */ | |
87 | if (((inst >> 16) & 0x1f) != 1) | |
88 | return 0; | |
89 | /* check major opcode */ | |
90 | switch (inst >> 26) { | |
91 | case 37: /* stwu */ | |
92 | case 39: /* stbu */ | |
93 | case 45: /* sthu */ | |
94 | case 53: /* stfsu */ | |
95 | case 55: /* stfdu */ | |
96 | return 1; | |
97 | case 62: /* std or stdu */ | |
98 | return (inst & 3) == 1; | |
99 | case 31: | |
100 | /* check minor opcode */ | |
101 | switch ((inst >> 1) & 0x3ff) { | |
102 | case 181: /* stdux */ | |
103 | case 183: /* stwux */ | |
104 | case 247: /* stbux */ | |
105 | case 439: /* sthux */ | |
106 | case 695: /* stfsux */ | |
107 | case 759: /* stfdux */ | |
108 | return 1; | |
109 | } | |
110 | } | |
111 | return 0; | |
112 | } | |
113 | ||
cffb09ce | 114 | #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) |
bce6c5fd AB |
115 | static void do_dabr(struct pt_regs *regs, unsigned long address, |
116 | unsigned long error_code) | |
14cf11af PM |
117 | { |
118 | siginfo_t info; | |
119 | ||
120 | if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code, | |
121 | 11, SIGSEGV) == NOTIFY_STOP) | |
122 | return; | |
123 | ||
124 | if (debugger_dabr_match(regs)) | |
125 | return; | |
126 | ||
127 | /* Clear the DABR */ | |
128 | set_dabr(0); | |
129 | ||
130 | /* Deliver the signal to userspace */ | |
131 | info.si_signo = SIGTRAP; | |
132 | info.si_errno = 0; | |
133 | info.si_code = TRAP_HWBKPT; | |
bce6c5fd | 134 | info.si_addr = (void __user *)address; |
14cf11af PM |
135 | force_sig_info(SIGTRAP, &info, current); |
136 | } | |
cffb09ce | 137 | #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/ |
14cf11af PM |
138 | |
139 | /* | |
140 | * For 600- and 800-family processors, the error_code parameter is DSISR | |
141 | * for a data fault, SRR1 for an instruction fault. For 400-family processors | |
142 | * the error_code parameter is ESR for a data fault, 0 for an instruction | |
143 | * fault. | |
144 | * For 64-bit processors, the error_code parameter is | |
145 | * - DSISR for a non-SLB data access fault, | |
146 | * - SRR1 & 0x08000000 for a non-SLB instruction access fault | |
147 | * - 0 any SLB fault. | |
148 | * | |
149 | * The return value is 0 if the fault was handled, or the signal | |
150 | * number if this is a kernel fault that can't be handled here. | |
151 | */ | |
152 | int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address, | |
153 | unsigned long error_code) | |
154 | { | |
155 | struct vm_area_struct * vma; | |
156 | struct mm_struct *mm = current->mm; | |
157 | siginfo_t info; | |
158 | int code = SEGV_MAPERR; | |
159 | int is_write = 0; | |
160 | int trap = TRAP(regs); | |
161 | int is_exec = trap == 0x400; | |
162 | ||
163 | #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) | |
164 | /* | |
165 | * Fortunately the bit assignments in SRR1 for an instruction | |
166 | * fault and DSISR for a data fault are mostly the same for the | |
167 | * bits we are interested in. But there are some bits which | |
168 | * indicate errors in DSISR but can validly be set in SRR1. | |
169 | */ | |
170 | if (trap == 0x400) | |
171 | error_code &= 0x48200000; | |
172 | else | |
173 | is_write = error_code & DSISR_ISSTORE; | |
174 | #else | |
175 | is_write = error_code & ESR_DST; | |
176 | #endif /* CONFIG_4xx || CONFIG_BOOKE */ | |
177 | ||
4f9e87c0 | 178 | if (notify_page_fault(DIE_PAGE_FAULT, "page_fault", regs, error_code, |
14cf11af PM |
179 | 11, SIGSEGV) == NOTIFY_STOP) |
180 | return 0; | |
181 | ||
182 | if (trap == 0x300) { | |
183 | if (debugger_fault_handler(regs)) | |
184 | return 0; | |
185 | } | |
186 | ||
187 | /* On a kernel SLB miss we can only check for a valid exception entry */ | |
188 | if (!user_mode(regs) && (address >= TASK_SIZE)) | |
189 | return SIGSEGV; | |
190 | ||
191 | #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) | |
192 | if (error_code & DSISR_DABRMATCH) { | |
193 | /* DABR match */ | |
bce6c5fd | 194 | do_dabr(regs, address, error_code); |
14cf11af PM |
195 | return 0; |
196 | } | |
197 | #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/ | |
198 | ||
199 | if (in_atomic() || mm == NULL) { | |
200 | if (!user_mode(regs)) | |
201 | return SIGSEGV; | |
202 | /* in_atomic() in user mode is really bad, | |
203 | as is current->mm == NULL. */ | |
204 | printk(KERN_EMERG "Page fault in user mode with" | |
205 | "in_atomic() = %d mm = %p\n", in_atomic(), mm); | |
206 | printk(KERN_EMERG "NIP = %lx MSR = %lx\n", | |
207 | regs->nip, regs->msr); | |
208 | die("Weird page fault", regs, SIGSEGV); | |
209 | } | |
210 | ||
211 | /* When running in the kernel we expect faults to occur only to | |
212 | * addresses in user space. All other faults represent errors in the | |
fc5266ea AB |
213 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
214 | * erroneous fault occurring in a code path which already holds mmap_sem | |
14cf11af PM |
215 | * we will deadlock attempting to validate the fault against the |
216 | * address space. Luckily the kernel only validly references user | |
217 | * space from well defined areas of code, which are listed in the | |
218 | * exceptions table. | |
219 | * | |
220 | * As the vast majority of faults will be valid we will only perform | |
fc5266ea | 221 | * the source reference check when there is a possibility of a deadlock. |
14cf11af PM |
222 | * Attempt to lock the address space, if we cannot we then validate the |
223 | * source. If this is invalid we can skip the address space check, | |
224 | * thus avoiding the deadlock. | |
225 | */ | |
226 | if (!down_read_trylock(&mm->mmap_sem)) { | |
227 | if (!user_mode(regs) && !search_exception_tables(regs->nip)) | |
228 | goto bad_area_nosemaphore; | |
229 | ||
230 | down_read(&mm->mmap_sem); | |
231 | } | |
232 | ||
233 | vma = find_vma(mm, address); | |
234 | if (!vma) | |
235 | goto bad_area; | |
236 | if (vma->vm_start <= address) | |
237 | goto good_area; | |
238 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
239 | goto bad_area; | |
240 | ||
241 | /* | |
242 | * N.B. The POWER/Open ABI allows programs to access up to | |
243 | * 288 bytes below the stack pointer. | |
244 | * The kernel signal delivery code writes up to about 1.5kB | |
245 | * below the stack pointer (r1) before decrementing it. | |
246 | * The exec code can write slightly over 640kB to the stack | |
247 | * before setting the user r1. Thus we allow the stack to | |
248 | * expand to 1MB without further checks. | |
249 | */ | |
250 | if (address + 0x100000 < vma->vm_end) { | |
251 | /* get user regs even if this fault is in kernel mode */ | |
252 | struct pt_regs *uregs = current->thread.regs; | |
253 | if (uregs == NULL) | |
254 | goto bad_area; | |
255 | ||
256 | /* | |
257 | * A user-mode access to an address a long way below | |
258 | * the stack pointer is only valid if the instruction | |
259 | * is one which would update the stack pointer to the | |
260 | * address accessed if the instruction completed, | |
261 | * i.e. either stwu rs,n(r1) or stwux rs,r1,rb | |
262 | * (or the byte, halfword, float or double forms). | |
263 | * | |
264 | * If we don't check this then any write to the area | |
265 | * between the last mapped region and the stack will | |
266 | * expand the stack rather than segfaulting. | |
267 | */ | |
268 | if (address + 2048 < uregs->gpr[1] | |
269 | && (!user_mode(regs) || !store_updates_sp(regs))) | |
270 | goto bad_area; | |
271 | } | |
272 | if (expand_stack(vma, address)) | |
273 | goto bad_area; | |
274 | ||
275 | good_area: | |
276 | code = SEGV_ACCERR; | |
277 | #if defined(CONFIG_6xx) | |
278 | if (error_code & 0x95700000) | |
279 | /* an error such as lwarx to I/O controller space, | |
280 | address matching DABR, eciwx, etc. */ | |
281 | goto bad_area; | |
282 | #endif /* CONFIG_6xx */ | |
283 | #if defined(CONFIG_8xx) | |
284 | /* The MPC8xx seems to always set 0x80000000, which is | |
285 | * "undefined". Of those that can be set, this is the only | |
286 | * one which seems bad. | |
287 | */ | |
288 | if (error_code & 0x10000000) | |
289 | /* Guarded storage error. */ | |
290 | goto bad_area; | |
291 | #endif /* CONFIG_8xx */ | |
292 | ||
293 | if (is_exec) { | |
294 | #ifdef CONFIG_PPC64 | |
295 | /* protection fault */ | |
296 | if (error_code & DSISR_PROTFAULT) | |
297 | goto bad_area; | |
298 | if (!(vma->vm_flags & VM_EXEC)) | |
299 | goto bad_area; | |
300 | #endif | |
301 | #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE) | |
302 | pte_t *ptep; | |
bab70a4a | 303 | pmd_t *pmdp; |
14cf11af PM |
304 | |
305 | /* Since 4xx/Book-E supports per-page execute permission, | |
306 | * we lazily flush dcache to icache. */ | |
307 | ptep = NULL; | |
bab70a4a ES |
308 | if (get_pteptr(mm, address, &ptep, &pmdp)) { |
309 | spinlock_t *ptl = pte_lockptr(mm, pmdp); | |
310 | spin_lock(ptl); | |
311 | if (pte_present(*ptep)) { | |
312 | struct page *page = pte_page(*ptep); | |
14cf11af | 313 | |
bab70a4a ES |
314 | if (!test_bit(PG_arch_1, &page->flags)) { |
315 | flush_dcache_icache_page(page); | |
316 | set_bit(PG_arch_1, &page->flags); | |
317 | } | |
318 | pte_update(ptep, 0, _PAGE_HWEXEC); | |
319 | _tlbie(address); | |
320 | pte_unmap_unlock(ptep, ptl); | |
321 | up_read(&mm->mmap_sem); | |
322 | return 0; | |
14cf11af | 323 | } |
bab70a4a | 324 | pte_unmap_unlock(ptep, ptl); |
14cf11af | 325 | } |
14cf11af PM |
326 | #endif |
327 | /* a write */ | |
328 | } else if (is_write) { | |
329 | if (!(vma->vm_flags & VM_WRITE)) | |
330 | goto bad_area; | |
331 | /* a read */ | |
332 | } else { | |
333 | /* protection fault */ | |
334 | if (error_code & 0x08000000) | |
335 | goto bad_area; | |
336 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) | |
337 | goto bad_area; | |
338 | } | |
339 | ||
340 | /* | |
341 | * If for any reason at all we couldn't handle the fault, | |
342 | * make sure we exit gracefully rather than endlessly redo | |
343 | * the fault. | |
344 | */ | |
345 | survive: | |
346 | switch (handle_mm_fault(mm, vma, address, is_write)) { | |
347 | ||
348 | case VM_FAULT_MINOR: | |
349 | current->min_flt++; | |
350 | break; | |
351 | case VM_FAULT_MAJOR: | |
352 | current->maj_flt++; | |
353 | break; | |
354 | case VM_FAULT_SIGBUS: | |
355 | goto do_sigbus; | |
356 | case VM_FAULT_OOM: | |
357 | goto out_of_memory; | |
358 | default: | |
359 | BUG(); | |
360 | } | |
361 | ||
362 | up_read(&mm->mmap_sem); | |
363 | return 0; | |
364 | ||
365 | bad_area: | |
366 | up_read(&mm->mmap_sem); | |
367 | ||
368 | bad_area_nosemaphore: | |
369 | /* User mode accesses cause a SIGSEGV */ | |
370 | if (user_mode(regs)) { | |
371 | _exception(SIGSEGV, regs, code, address); | |
372 | return 0; | |
373 | } | |
374 | ||
375 | if (is_exec && (error_code & DSISR_PROTFAULT) | |
376 | && printk_ratelimit()) | |
377 | printk(KERN_CRIT "kernel tried to execute NX-protected" | |
378 | " page (%lx) - exploit attempt? (uid: %d)\n", | |
379 | address, current->uid); | |
380 | ||
381 | return SIGSEGV; | |
382 | ||
383 | /* | |
384 | * We ran out of memory, or some other thing happened to us that made | |
385 | * us unable to handle the page fault gracefully. | |
386 | */ | |
387 | out_of_memory: | |
388 | up_read(&mm->mmap_sem); | |
389 | if (current->pid == 1) { | |
390 | yield(); | |
391 | down_read(&mm->mmap_sem); | |
392 | goto survive; | |
393 | } | |
394 | printk("VM: killing process %s\n", current->comm); | |
395 | if (user_mode(regs)) | |
396 | do_exit(SIGKILL); | |
397 | return SIGKILL; | |
398 | ||
399 | do_sigbus: | |
400 | up_read(&mm->mmap_sem); | |
401 | if (user_mode(regs)) { | |
402 | info.si_signo = SIGBUS; | |
403 | info.si_errno = 0; | |
404 | info.si_code = BUS_ADRERR; | |
405 | info.si_addr = (void __user *)address; | |
406 | force_sig_info(SIGBUS, &info, current); | |
407 | return 0; | |
408 | } | |
409 | return SIGBUS; | |
410 | } | |
411 | ||
412 | /* | |
413 | * bad_page_fault is called when we have a bad access from the kernel. | |
414 | * It is called from the DSI and ISI handlers in head.S and from some | |
415 | * of the procedures in traps.c. | |
416 | */ | |
417 | void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) | |
418 | { | |
419 | const struct exception_table_entry *entry; | |
420 | ||
421 | /* Are we prepared to handle this fault? */ | |
422 | if ((entry = search_exception_tables(regs->nip)) != NULL) { | |
423 | regs->nip = entry->fixup; | |
424 | return; | |
425 | } | |
426 | ||
427 | /* kernel has accessed a bad area */ | |
723925b7 OJ |
428 | |
429 | printk(KERN_ALERT "Unable to handle kernel paging request for "); | |
430 | switch (regs->trap) { | |
431 | case 0x300: | |
432 | case 0x380: | |
433 | printk("data at address 0x%08lx\n", regs->dar); | |
434 | break; | |
435 | case 0x400: | |
436 | case 0x480: | |
437 | printk("instruction fetch\n"); | |
438 | break; | |
439 | default: | |
440 | printk("unknown fault\n"); | |
441 | } | |
442 | printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", | |
443 | regs->nip); | |
444 | ||
14cf11af PM |
445 | die("Kernel access of bad area", regs, sig); |
446 | } |