Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * arch/s390/mm/fault.c | |
3 | * | |
4 | * S390 version | |
5 | * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation | |
6 | * Author(s): Hartmut Penner (hp@de.ibm.com) | |
7 | * Ulrich Weigand (uweigand@de.ibm.com) | |
8 | * | |
9 | * Derived from "arch/i386/mm/fault.c" | |
10 | * Copyright (C) 1995 Linus Torvalds | |
11 | */ | |
12 | ||
1da177e4 LT |
13 | #include <linux/signal.h> |
14 | #include <linux/sched.h> | |
15 | #include <linux/kernel.h> | |
16 | #include <linux/errno.h> | |
17 | #include <linux/string.h> | |
18 | #include <linux/types.h> | |
19 | #include <linux/ptrace.h> | |
20 | #include <linux/mman.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/smp.h> | |
23 | #include <linux/smp_lock.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/console.h> | |
26 | #include <linux/module.h> | |
27 | #include <linux/hardirq.h> | |
4ba069b8 | 28 | #include <linux/kprobes.h> |
1da177e4 LT |
29 | |
30 | #include <asm/system.h> | |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/pgtable.h> | |
4ba069b8 | 33 | #include <asm/kdebug.h> |
1da177e4 | 34 | |
347a8dc3 | 35 | #ifndef CONFIG_64BIT |
1da177e4 LT |
36 | #define __FAIL_ADDR_MASK 0x7ffff000 |
37 | #define __FIXUP_MASK 0x7fffffff | |
38 | #define __SUBCODE_MASK 0x0200 | |
39 | #define __PF_RES_FIELD 0ULL | |
347a8dc3 | 40 | #else /* CONFIG_64BIT */ |
1da177e4 LT |
41 | #define __FAIL_ADDR_MASK -4096L |
42 | #define __FIXUP_MASK ~0L | |
43 | #define __SUBCODE_MASK 0x0600 | |
44 | #define __PF_RES_FIELD 0x8000000000000000ULL | |
347a8dc3 | 45 | #endif /* CONFIG_64BIT */ |
1da177e4 LT |
46 | |
47 | #ifdef CONFIG_SYSCTL | |
48 | extern int sysctl_userprocess_debug; | |
49 | #endif | |
50 | ||
51 | extern void die(const char *,struct pt_regs *,long); | |
52 | ||
4ba069b8 MG |
53 | #ifdef CONFIG_KPROBES |
54 | ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain); | |
55 | int register_page_fault_notifier(struct notifier_block *nb) | |
56 | { | |
57 | return atomic_notifier_chain_register(¬ify_page_fault_chain, nb); | |
58 | } | |
59 | ||
60 | int unregister_page_fault_notifier(struct notifier_block *nb) | |
61 | { | |
62 | return atomic_notifier_chain_unregister(¬ify_page_fault_chain, nb); | |
63 | } | |
64 | ||
65 | static inline int notify_page_fault(enum die_val val, const char *str, | |
66 | struct pt_regs *regs, long err, int trap, int sig) | |
67 | { | |
68 | struct die_args args = { | |
69 | .regs = regs, | |
70 | .str = str, | |
71 | .err = err, | |
72 | .trapnr = trap, | |
73 | .signr = sig | |
74 | }; | |
75 | return atomic_notifier_call_chain(¬ify_page_fault_chain, val, &args); | |
76 | } | |
77 | #else | |
78 | static inline int notify_page_fault(enum die_val val, const char *str, | |
79 | struct pt_regs *regs, long err, int trap, int sig) | |
80 | { | |
81 | return NOTIFY_DONE; | |
82 | } | |
83 | #endif | |
84 | ||
1da177e4 LT |
85 | extern spinlock_t timerlist_lock; |
86 | ||
87 | /* | |
88 | * Unlock any spinlocks which will prevent us from getting the | |
89 | * message out (timerlist_lock is acquired through the | |
90 | * console unblank code) | |
91 | */ | |
92 | void bust_spinlocks(int yes) | |
93 | { | |
94 | if (yes) { | |
95 | oops_in_progress = 1; | |
96 | } else { | |
97 | int loglevel_save = console_loglevel; | |
98 | console_unblank(); | |
99 | oops_in_progress = 0; | |
100 | /* | |
101 | * OK, the message is on the console. Now we call printk() | |
102 | * without oops_in_progress set so that printk will give klogd | |
103 | * a poke. Hold onto your hats... | |
104 | */ | |
105 | console_loglevel = 15; | |
106 | printk(" "); | |
107 | console_loglevel = loglevel_save; | |
108 | } | |
109 | } | |
110 | ||
111 | /* | |
112 | * Check which address space is addressed by the access | |
113 | * register in S390_lowcore.exc_access_id. | |
114 | * Returns 1 for user space and 0 for kernel space. | |
115 | */ | |
116 | static int __check_access_register(struct pt_regs *regs, int error_code) | |
117 | { | |
118 | int areg = S390_lowcore.exc_access_id; | |
119 | ||
120 | if (areg == 0) | |
121 | /* Access via access register 0 -> kernel address */ | |
122 | return 0; | |
123 | save_access_regs(current->thread.acrs); | |
124 | if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1) | |
125 | /* | |
126 | * access register contains 0 -> kernel address, | |
127 | * access register contains 1 -> user space address | |
128 | */ | |
129 | return current->thread.acrs[areg]; | |
130 | ||
131 | /* Something unhealthy was done with the access registers... */ | |
132 | die("page fault via unknown access register", regs, error_code); | |
133 | do_exit(SIGKILL); | |
134 | return 0; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Check which address space the address belongs to. | |
139 | * Returns 1 for user space and 0 for kernel space. | |
140 | */ | |
141 | static inline int check_user_space(struct pt_regs *regs, int error_code) | |
142 | { | |
143 | /* | |
144 | * The lowest two bits of S390_lowcore.trans_exc_code indicate | |
145 | * which paging table was used: | |
146 | * 0: Primary Segment Table Descriptor | |
147 | * 1: STD determined via access register | |
148 | * 2: Secondary Segment Table Descriptor | |
149 | * 3: Home Segment Table Descriptor | |
150 | */ | |
151 | int descriptor = S390_lowcore.trans_exc_code & 3; | |
152 | if (unlikely(descriptor == 1)) | |
153 | return __check_access_register(regs, error_code); | |
154 | if (descriptor == 2) | |
155 | return current->thread.mm_segment.ar4; | |
156 | return descriptor != 0; | |
157 | } | |
158 | ||
159 | /* | |
160 | * Send SIGSEGV to task. This is an external routine | |
161 | * to keep the stack usage of do_page_fault small. | |
162 | */ | |
163 | static void do_sigsegv(struct pt_regs *regs, unsigned long error_code, | |
164 | int si_code, unsigned long address) | |
165 | { | |
166 | struct siginfo si; | |
167 | ||
168 | #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG) | |
169 | #if defined(CONFIG_SYSCTL) | |
170 | if (sysctl_userprocess_debug) | |
171 | #endif | |
172 | { | |
173 | printk("User process fault: interruption code 0x%lX\n", | |
174 | error_code); | |
175 | printk("failing address: %lX\n", address); | |
176 | show_regs(regs); | |
177 | } | |
178 | #endif | |
179 | si.si_signo = SIGSEGV; | |
180 | si.si_code = si_code; | |
d2c993d8 | 181 | si.si_addr = (void __user *) address; |
1da177e4 LT |
182 | force_sig_info(SIGSEGV, &si, current); |
183 | } | |
184 | ||
185 | /* | |
186 | * This routine handles page faults. It determines the address, | |
187 | * and the problem, and then passes it off to one of the appropriate | |
188 | * routines. | |
189 | * | |
190 | * error_code: | |
191 | * 04 Protection -> Write-Protection (suprression) | |
192 | * 10 Segment translation -> Not present (nullification) | |
193 | * 11 Page translation -> Not present (nullification) | |
194 | * 3b Region third trans. -> Not present (nullification) | |
195 | */ | |
4ba069b8 | 196 | static inline void __kprobes |
1da177e4 LT |
197 | do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection) |
198 | { | |
199 | struct task_struct *tsk; | |
200 | struct mm_struct *mm; | |
201 | struct vm_area_struct * vma; | |
202 | unsigned long address; | |
203 | int user_address; | |
204 | const struct exception_table_entry *fixup; | |
205 | int si_code = SEGV_MAPERR; | |
206 | ||
207 | tsk = current; | |
208 | mm = tsk->mm; | |
209 | ||
4ba069b8 MG |
210 | if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14, |
211 | SIGSEGV) == NOTIFY_STOP) | |
212 | return; | |
213 | ||
1da177e4 LT |
214 | /* |
215 | * Check for low-address protection. This needs to be treated | |
216 | * as a special case because the translation exception code | |
217 | * field is not guaranteed to contain valid data in this case. | |
218 | */ | |
219 | if (is_protection && !(S390_lowcore.trans_exc_code & 4)) { | |
220 | ||
221 | /* Low-address protection hit in kernel mode means | |
222 | NULL pointer write access in kernel mode. */ | |
223 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) { | |
224 | address = 0; | |
225 | user_address = 0; | |
226 | goto no_context; | |
227 | } | |
228 | ||
229 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
230 | die ("Low-address protection", regs, error_code); | |
231 | do_exit(SIGKILL); | |
232 | } | |
233 | ||
234 | /* | |
235 | * get the failing address | |
236 | * more specific the segment and page table portion of | |
237 | * the address | |
238 | */ | |
239 | address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK; | |
240 | user_address = check_user_space(regs, error_code); | |
241 | ||
242 | /* | |
243 | * Verify that the fault happened in user space, that | |
244 | * we are not in an interrupt and that there is a | |
245 | * user context. | |
246 | */ | |
595bf2aa | 247 | if (user_address == 0 || in_atomic() || !mm) |
1da177e4 LT |
248 | goto no_context; |
249 | ||
250 | /* | |
251 | * When we get here, the fault happened in the current | |
252 | * task's user address space, so we can switch on the | |
253 | * interrupts again and then search the VMAs | |
254 | */ | |
255 | local_irq_enable(); | |
256 | ||
257 | down_read(&mm->mmap_sem); | |
258 | ||
259 | vma = find_vma(mm, address); | |
260 | if (!vma) | |
261 | goto bad_area; | |
262 | if (vma->vm_start <= address) | |
263 | goto good_area; | |
264 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
265 | goto bad_area; | |
266 | if (expand_stack(vma, address)) | |
267 | goto bad_area; | |
268 | /* | |
269 | * Ok, we have a good vm_area for this memory access, so | |
270 | * we can handle it.. | |
271 | */ | |
272 | good_area: | |
273 | si_code = SEGV_ACCERR; | |
274 | if (!is_protection) { | |
275 | /* page not present, check vm flags */ | |
276 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) | |
277 | goto bad_area; | |
278 | } else { | |
279 | if (!(vma->vm_flags & VM_WRITE)) | |
280 | goto bad_area; | |
281 | } | |
282 | ||
283 | survive: | |
284 | /* | |
285 | * If for any reason at all we couldn't handle the fault, | |
286 | * make sure we exit gracefully rather than endlessly redo | |
287 | * the fault. | |
288 | */ | |
289 | switch (handle_mm_fault(mm, vma, address, is_protection)) { | |
290 | case VM_FAULT_MINOR: | |
291 | tsk->min_flt++; | |
292 | break; | |
293 | case VM_FAULT_MAJOR: | |
294 | tsk->maj_flt++; | |
295 | break; | |
296 | case VM_FAULT_SIGBUS: | |
297 | goto do_sigbus; | |
298 | case VM_FAULT_OOM: | |
299 | goto out_of_memory; | |
300 | default: | |
301 | BUG(); | |
302 | } | |
303 | ||
304 | up_read(&mm->mmap_sem); | |
305 | /* | |
306 | * The instruction that caused the program check will | |
307 | * be repeated. Don't signal single step via SIGTRAP. | |
308 | */ | |
309 | clear_tsk_thread_flag(current, TIF_SINGLE_STEP); | |
310 | return; | |
311 | ||
312 | /* | |
313 | * Something tried to access memory that isn't in our memory map.. | |
314 | * Fix it, but check if it's kernel or user first.. | |
315 | */ | |
316 | bad_area: | |
317 | up_read(&mm->mmap_sem); | |
318 | ||
319 | /* User mode accesses just cause a SIGSEGV */ | |
320 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
321 | tsk->thread.prot_addr = address; | |
322 | tsk->thread.trap_no = error_code; | |
323 | do_sigsegv(regs, error_code, si_code, address); | |
324 | return; | |
325 | } | |
326 | ||
327 | no_context: | |
328 | /* Are we prepared to handle this kernel fault? */ | |
329 | fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK); | |
330 | if (fixup) { | |
331 | regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE; | |
332 | return; | |
333 | } | |
334 | ||
335 | /* | |
336 | * Oops. The kernel tried to access some bad page. We'll have to | |
337 | * terminate things with extreme prejudice. | |
338 | */ | |
339 | if (user_address == 0) | |
340 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" | |
341 | " at virtual kernel address %p\n", (void *)address); | |
342 | else | |
343 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
344 | " at virtual user address %p\n", (void *)address); | |
345 | ||
346 | die("Oops", regs, error_code); | |
347 | do_exit(SIGKILL); | |
348 | ||
349 | ||
350 | /* | |
351 | * We ran out of memory, or some other thing happened to us that made | |
352 | * us unable to handle the page fault gracefully. | |
353 | */ | |
354 | out_of_memory: | |
355 | up_read(&mm->mmap_sem); | |
356 | if (tsk->pid == 1) { | |
357 | yield(); | |
bac9c66c | 358 | down_read(&mm->mmap_sem); |
1da177e4 LT |
359 | goto survive; |
360 | } | |
361 | printk("VM: killing process %s\n", tsk->comm); | |
362 | if (regs->psw.mask & PSW_MASK_PSTATE) | |
363 | do_exit(SIGKILL); | |
364 | goto no_context; | |
365 | ||
366 | do_sigbus: | |
367 | up_read(&mm->mmap_sem); | |
368 | ||
369 | /* | |
370 | * Send a sigbus, regardless of whether we were in kernel | |
371 | * or user mode. | |
372 | */ | |
373 | tsk->thread.prot_addr = address; | |
374 | tsk->thread.trap_no = error_code; | |
375 | force_sig(SIGBUS, tsk); | |
376 | ||
377 | /* Kernel mode? Handle exceptions or die */ | |
378 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
379 | goto no_context; | |
380 | } | |
381 | ||
382 | void do_protection_exception(struct pt_regs *regs, unsigned long error_code) | |
383 | { | |
384 | regs->psw.addr -= (error_code >> 16); | |
385 | do_exception(regs, 4, 1); | |
386 | } | |
387 | ||
388 | void do_dat_exception(struct pt_regs *regs, unsigned long error_code) | |
389 | { | |
390 | do_exception(regs, error_code & 0xff, 0); | |
391 | } | |
392 | ||
1da177e4 LT |
393 | #ifdef CONFIG_PFAULT |
394 | /* | |
395 | * 'pfault' pseudo page faults routines. | |
396 | */ | |
397 | static int pfault_disable = 0; | |
398 | ||
399 | static int __init nopfault(char *str) | |
400 | { | |
401 | pfault_disable = 1; | |
402 | return 1; | |
403 | } | |
404 | ||
405 | __setup("nopfault", nopfault); | |
406 | ||
407 | typedef struct { | |
408 | __u16 refdiagc; | |
409 | __u16 reffcode; | |
410 | __u16 refdwlen; | |
411 | __u16 refversn; | |
412 | __u64 refgaddr; | |
413 | __u64 refselmk; | |
414 | __u64 refcmpmk; | |
415 | __u64 reserved; | |
416 | } __attribute__ ((packed)) pfault_refbk_t; | |
417 | ||
418 | int pfault_init(void) | |
419 | { | |
420 | pfault_refbk_t refbk = | |
421 | { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48, | |
422 | __PF_RES_FIELD }; | |
423 | int rc; | |
424 | ||
425 | if (pfault_disable) | |
426 | return -1; | |
94c12cc7 MS |
427 | asm volatile( |
428 | " diag %1,%0,0x258\n" | |
429 | "0: j 2f\n" | |
430 | "1: la %0,8\n" | |
1da177e4 | 431 | "2:\n" |
94c12cc7 MS |
432 | EX_TABLE(0b,1b) |
433 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
434 | __ctl_set_bit(0, 9); |
435 | return rc; | |
436 | } | |
437 | ||
438 | void pfault_fini(void) | |
439 | { | |
440 | pfault_refbk_t refbk = | |
441 | { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL }; | |
442 | ||
443 | if (pfault_disable) | |
444 | return; | |
445 | __ctl_clear_bit(0,9); | |
94c12cc7 MS |
446 | asm volatile( |
447 | " diag %0,0,0x258\n" | |
1da177e4 | 448 | "0:\n" |
94c12cc7 MS |
449 | EX_TABLE(0b,0b) |
450 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
451 | } |
452 | ||
453 | asmlinkage void | |
454 | pfault_interrupt(struct pt_regs *regs, __u16 error_code) | |
455 | { | |
456 | struct task_struct *tsk; | |
457 | __u16 subcode; | |
458 | ||
459 | /* | |
460 | * Get the external interruption subcode & pfault | |
461 | * initial/completion signal bit. VM stores this | |
462 | * in the 'cpu address' field associated with the | |
463 | * external interrupt. | |
464 | */ | |
465 | subcode = S390_lowcore.cpu_addr; | |
466 | if ((subcode & 0xff00) != __SUBCODE_MASK) | |
467 | return; | |
468 | ||
469 | /* | |
470 | * Get the token (= address of the task structure of the affected task). | |
471 | */ | |
472 | tsk = *(struct task_struct **) __LC_PFAULT_INTPARM; | |
473 | ||
474 | if (subcode & 0x0080) { | |
475 | /* signal bit is set -> a page has been swapped in by VM */ | |
476 | if (xchg(&tsk->thread.pfault_wait, -1) != 0) { | |
477 | /* Initial interrupt was faster than the completion | |
478 | * interrupt. pfault_wait is valid. Set pfault_wait | |
479 | * back to zero and wake up the process. This can | |
480 | * safely be done because the task is still sleeping | |
b6d09449 | 481 | * and can't produce new pfaults. */ |
1da177e4 LT |
482 | tsk->thread.pfault_wait = 0; |
483 | wake_up_process(tsk); | |
b6d09449 | 484 | put_task_struct(tsk); |
1da177e4 LT |
485 | } |
486 | } else { | |
487 | /* signal bit not set -> a real page is missing. */ | |
b6d09449 | 488 | get_task_struct(tsk); |
1da177e4 LT |
489 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
490 | if (xchg(&tsk->thread.pfault_wait, 1) != 0) { | |
491 | /* Completion interrupt was faster than the initial | |
492 | * interrupt (swapped in a -1 for pfault_wait). Set | |
493 | * pfault_wait back to zero and exit. This can be | |
494 | * done safely because tsk is running in kernel | |
495 | * mode and can't produce new pfaults. */ | |
496 | tsk->thread.pfault_wait = 0; | |
497 | set_task_state(tsk, TASK_RUNNING); | |
b6d09449 | 498 | put_task_struct(tsk); |
1da177e4 LT |
499 | } else |
500 | set_tsk_need_resched(tsk); | |
501 | } | |
502 | } | |
503 | #endif | |
504 |