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> |
29b08d2b | 34 | #include <asm/s390_ext.h> |
1da177e4 | 35 | |
347a8dc3 | 36 | #ifndef CONFIG_64BIT |
1da177e4 LT |
37 | #define __FAIL_ADDR_MASK 0x7ffff000 |
38 | #define __FIXUP_MASK 0x7fffffff | |
39 | #define __SUBCODE_MASK 0x0200 | |
40 | #define __PF_RES_FIELD 0ULL | |
347a8dc3 | 41 | #else /* CONFIG_64BIT */ |
1da177e4 LT |
42 | #define __FAIL_ADDR_MASK -4096L |
43 | #define __FIXUP_MASK ~0L | |
44 | #define __SUBCODE_MASK 0x0600 | |
45 | #define __PF_RES_FIELD 0x8000000000000000ULL | |
347a8dc3 | 46 | #endif /* CONFIG_64BIT */ |
1da177e4 LT |
47 | |
48 | #ifdef CONFIG_SYSCTL | |
49 | extern int sysctl_userprocess_debug; | |
50 | #endif | |
51 | ||
52 | extern void die(const char *,struct pt_regs *,long); | |
53 | ||
4ba069b8 | 54 | #ifdef CONFIG_KPROBES |
2b67fc46 | 55 | static ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain); |
4ba069b8 MG |
56 | int register_page_fault_notifier(struct notifier_block *nb) |
57 | { | |
58 | return atomic_notifier_chain_register(¬ify_page_fault_chain, nb); | |
59 | } | |
60 | ||
61 | int unregister_page_fault_notifier(struct notifier_block *nb) | |
62 | { | |
63 | return atomic_notifier_chain_unregister(¬ify_page_fault_chain, nb); | |
64 | } | |
65 | ||
66 | static inline int notify_page_fault(enum die_val val, const char *str, | |
67 | struct pt_regs *regs, long err, int trap, int sig) | |
68 | { | |
69 | struct die_args args = { | |
70 | .regs = regs, | |
71 | .str = str, | |
72 | .err = err, | |
73 | .trapnr = trap, | |
74 | .signr = sig | |
75 | }; | |
76 | return atomic_notifier_call_chain(¬ify_page_fault_chain, val, &args); | |
77 | } | |
78 | #else | |
79 | static inline int notify_page_fault(enum die_val val, const char *str, | |
80 | struct pt_regs *regs, long err, int trap, int sig) | |
81 | { | |
82 | return NOTIFY_DONE; | |
83 | } | |
84 | #endif | |
85 | ||
1da177e4 LT |
86 | |
87 | /* | |
88 | * Unlock any spinlocks which will prevent us from getting the | |
cefc8be8 | 89 | * message out. |
1da177e4 LT |
90 | */ |
91 | void bust_spinlocks(int yes) | |
92 | { | |
93 | if (yes) { | |
94 | oops_in_progress = 1; | |
95 | } else { | |
96 | int loglevel_save = console_loglevel; | |
97 | console_unblank(); | |
98 | oops_in_progress = 0; | |
99 | /* | |
100 | * OK, the message is on the console. Now we call printk() | |
101 | * without oops_in_progress set so that printk will give klogd | |
102 | * a poke. Hold onto your hats... | |
103 | */ | |
104 | console_loglevel = 15; | |
105 | printk(" "); | |
106 | console_loglevel = loglevel_save; | |
107 | } | |
108 | } | |
109 | ||
110 | /* | |
111 | * Check which address space is addressed by the access | |
112 | * register in S390_lowcore.exc_access_id. | |
113 | * Returns 1 for user space and 0 for kernel space. | |
114 | */ | |
115 | static int __check_access_register(struct pt_regs *regs, int error_code) | |
116 | { | |
117 | int areg = S390_lowcore.exc_access_id; | |
118 | ||
119 | if (areg == 0) | |
120 | /* Access via access register 0 -> kernel address */ | |
121 | return 0; | |
122 | save_access_regs(current->thread.acrs); | |
123 | if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1) | |
124 | /* | |
125 | * access register contains 0 -> kernel address, | |
126 | * access register contains 1 -> user space address | |
127 | */ | |
128 | return current->thread.acrs[areg]; | |
129 | ||
130 | /* Something unhealthy was done with the access registers... */ | |
131 | die("page fault via unknown access register", regs, error_code); | |
132 | do_exit(SIGKILL); | |
133 | return 0; | |
134 | } | |
135 | ||
136 | /* | |
137 | * Check which address space the address belongs to. | |
c1821c2e GS |
138 | * May return 1 or 2 for user space and 0 for kernel space. |
139 | * Returns 2 for user space in primary addressing mode with | |
140 | * CONFIG_S390_EXEC_PROTECT on and kernel parameter noexec=on. | |
1da177e4 LT |
141 | */ |
142 | static inline int check_user_space(struct pt_regs *regs, int error_code) | |
143 | { | |
144 | /* | |
145 | * The lowest two bits of S390_lowcore.trans_exc_code indicate | |
146 | * which paging table was used: | |
147 | * 0: Primary Segment Table Descriptor | |
148 | * 1: STD determined via access register | |
149 | * 2: Secondary Segment Table Descriptor | |
150 | * 3: Home Segment Table Descriptor | |
151 | */ | |
152 | int descriptor = S390_lowcore.trans_exc_code & 3; | |
153 | if (unlikely(descriptor == 1)) | |
154 | return __check_access_register(regs, error_code); | |
155 | if (descriptor == 2) | |
156 | return current->thread.mm_segment.ar4; | |
c1821c2e | 157 | return ((descriptor != 0) ^ (switch_amode)) << s390_noexec; |
1da177e4 LT |
158 | } |
159 | ||
160 | /* | |
161 | * Send SIGSEGV to task. This is an external routine | |
162 | * to keep the stack usage of do_page_fault small. | |
163 | */ | |
164 | static void do_sigsegv(struct pt_regs *regs, unsigned long error_code, | |
165 | int si_code, unsigned long address) | |
166 | { | |
167 | struct siginfo si; | |
168 | ||
169 | #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG) | |
170 | #if defined(CONFIG_SYSCTL) | |
171 | if (sysctl_userprocess_debug) | |
172 | #endif | |
173 | { | |
174 | printk("User process fault: interruption code 0x%lX\n", | |
175 | error_code); | |
176 | printk("failing address: %lX\n", address); | |
177 | show_regs(regs); | |
178 | } | |
179 | #endif | |
180 | si.si_signo = SIGSEGV; | |
181 | si.si_code = si_code; | |
d2c993d8 | 182 | si.si_addr = (void __user *) address; |
1da177e4 LT |
183 | force_sig_info(SIGSEGV, &si, current); |
184 | } | |
185 | ||
c1821c2e GS |
186 | #ifdef CONFIG_S390_EXEC_PROTECT |
187 | extern long sys_sigreturn(struct pt_regs *regs); | |
188 | extern long sys_rt_sigreturn(struct pt_regs *regs); | |
189 | extern long sys32_sigreturn(struct pt_regs *regs); | |
190 | extern long sys32_rt_sigreturn(struct pt_regs *regs); | |
191 | ||
192 | static inline void do_sigreturn(struct mm_struct *mm, struct pt_regs *regs, | |
193 | int rt) | |
194 | { | |
195 | up_read(&mm->mmap_sem); | |
196 | clear_tsk_thread_flag(current, TIF_SINGLE_STEP); | |
197 | #ifdef CONFIG_COMPAT | |
198 | if (test_tsk_thread_flag(current, TIF_31BIT)) { | |
199 | if (rt) | |
200 | sys32_rt_sigreturn(regs); | |
201 | else | |
202 | sys32_sigreturn(regs); | |
203 | return; | |
204 | } | |
205 | #endif /* CONFIG_COMPAT */ | |
206 | if (rt) | |
207 | sys_rt_sigreturn(regs); | |
208 | else | |
209 | sys_sigreturn(regs); | |
210 | return; | |
211 | } | |
212 | ||
213 | static int signal_return(struct mm_struct *mm, struct pt_regs *regs, | |
214 | unsigned long address, unsigned long error_code) | |
215 | { | |
216 | pgd_t *pgd; | |
217 | pmd_t *pmd; | |
218 | pte_t *pte; | |
219 | u16 *instruction; | |
220 | unsigned long pfn, uaddr = regs->psw.addr; | |
221 | ||
222 | spin_lock(&mm->page_table_lock); | |
223 | pgd = pgd_offset(mm, uaddr); | |
224 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
225 | goto out_fault; | |
226 | pmd = pmd_offset(pgd, uaddr); | |
227 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
228 | goto out_fault; | |
229 | pte = pte_offset_map(pmd_offset(pgd_offset(mm, uaddr), uaddr), uaddr); | |
230 | if (!pte || !pte_present(*pte)) | |
231 | goto out_fault; | |
232 | pfn = pte_pfn(*pte); | |
233 | if (!pfn_valid(pfn)) | |
234 | goto out_fault; | |
235 | spin_unlock(&mm->page_table_lock); | |
236 | ||
237 | instruction = (u16 *) ((pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE-1))); | |
238 | if (*instruction == 0x0a77) | |
239 | do_sigreturn(mm, regs, 0); | |
240 | else if (*instruction == 0x0aad) | |
241 | do_sigreturn(mm, regs, 1); | |
242 | else { | |
243 | printk("- XXX - do_exception: task = %s, primary, NO EXEC " | |
244 | "-> SIGSEGV\n", current->comm); | |
245 | up_read(&mm->mmap_sem); | |
246 | current->thread.prot_addr = address; | |
247 | current->thread.trap_no = error_code; | |
248 | do_sigsegv(regs, error_code, SEGV_MAPERR, address); | |
249 | } | |
250 | return 0; | |
251 | out_fault: | |
252 | spin_unlock(&mm->page_table_lock); | |
253 | return -EFAULT; | |
254 | } | |
255 | #endif /* CONFIG_S390_EXEC_PROTECT */ | |
256 | ||
1da177e4 LT |
257 | /* |
258 | * This routine handles page faults. It determines the address, | |
259 | * and the problem, and then passes it off to one of the appropriate | |
260 | * routines. | |
261 | * | |
262 | * error_code: | |
263 | * 04 Protection -> Write-Protection (suprression) | |
264 | * 10 Segment translation -> Not present (nullification) | |
265 | * 11 Page translation -> Not present (nullification) | |
266 | * 3b Region third trans. -> Not present (nullification) | |
267 | */ | |
4ba069b8 | 268 | static inline void __kprobes |
1da177e4 LT |
269 | do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection) |
270 | { | |
271 | struct task_struct *tsk; | |
272 | struct mm_struct *mm; | |
273 | struct vm_area_struct * vma; | |
274 | unsigned long address; | |
275 | int user_address; | |
276 | const struct exception_table_entry *fixup; | |
277 | int si_code = SEGV_MAPERR; | |
278 | ||
279 | tsk = current; | |
280 | mm = tsk->mm; | |
281 | ||
4ba069b8 MG |
282 | if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14, |
283 | SIGSEGV) == NOTIFY_STOP) | |
284 | return; | |
285 | ||
1da177e4 LT |
286 | /* |
287 | * Check for low-address protection. This needs to be treated | |
288 | * as a special case because the translation exception code | |
289 | * field is not guaranteed to contain valid data in this case. | |
290 | */ | |
291 | if (is_protection && !(S390_lowcore.trans_exc_code & 4)) { | |
292 | ||
293 | /* Low-address protection hit in kernel mode means | |
294 | NULL pointer write access in kernel mode. */ | |
295 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) { | |
296 | address = 0; | |
297 | user_address = 0; | |
298 | goto no_context; | |
299 | } | |
300 | ||
301 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
302 | die ("Low-address protection", regs, error_code); | |
303 | do_exit(SIGKILL); | |
304 | } | |
305 | ||
306 | /* | |
307 | * get the failing address | |
308 | * more specific the segment and page table portion of | |
309 | * the address | |
310 | */ | |
311 | address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK; | |
312 | user_address = check_user_space(regs, error_code); | |
313 | ||
314 | /* | |
315 | * Verify that the fault happened in user space, that | |
316 | * we are not in an interrupt and that there is a | |
317 | * user context. | |
318 | */ | |
595bf2aa | 319 | if (user_address == 0 || in_atomic() || !mm) |
1da177e4 LT |
320 | goto no_context; |
321 | ||
322 | /* | |
323 | * When we get here, the fault happened in the current | |
324 | * task's user address space, so we can switch on the | |
325 | * interrupts again and then search the VMAs | |
326 | */ | |
327 | local_irq_enable(); | |
328 | ||
329 | down_read(&mm->mmap_sem); | |
330 | ||
331 | vma = find_vma(mm, address); | |
332 | if (!vma) | |
333 | goto bad_area; | |
c1821c2e GS |
334 | |
335 | #ifdef CONFIG_S390_EXEC_PROTECT | |
336 | if (unlikely((user_address == 2) && !(vma->vm_flags & VM_EXEC))) | |
337 | if (!signal_return(mm, regs, address, error_code)) | |
338 | /* | |
339 | * signal_return() has done an up_read(&mm->mmap_sem) | |
340 | * if it returns 0. | |
341 | */ | |
342 | return; | |
343 | #endif | |
344 | ||
1da177e4 LT |
345 | if (vma->vm_start <= address) |
346 | goto good_area; | |
347 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
348 | goto bad_area; | |
349 | if (expand_stack(vma, address)) | |
350 | goto bad_area; | |
351 | /* | |
352 | * Ok, we have a good vm_area for this memory access, so | |
353 | * we can handle it.. | |
354 | */ | |
355 | good_area: | |
356 | si_code = SEGV_ACCERR; | |
357 | if (!is_protection) { | |
358 | /* page not present, check vm flags */ | |
359 | if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) | |
360 | goto bad_area; | |
361 | } else { | |
362 | if (!(vma->vm_flags & VM_WRITE)) | |
363 | goto bad_area; | |
364 | } | |
365 | ||
366 | survive: | |
367 | /* | |
368 | * If for any reason at all we couldn't handle the fault, | |
369 | * make sure we exit gracefully rather than endlessly redo | |
370 | * the fault. | |
371 | */ | |
372 | switch (handle_mm_fault(mm, vma, address, is_protection)) { | |
373 | case VM_FAULT_MINOR: | |
374 | tsk->min_flt++; | |
375 | break; | |
376 | case VM_FAULT_MAJOR: | |
377 | tsk->maj_flt++; | |
378 | break; | |
379 | case VM_FAULT_SIGBUS: | |
380 | goto do_sigbus; | |
381 | case VM_FAULT_OOM: | |
382 | goto out_of_memory; | |
383 | default: | |
384 | BUG(); | |
385 | } | |
386 | ||
387 | up_read(&mm->mmap_sem); | |
388 | /* | |
389 | * The instruction that caused the program check will | |
390 | * be repeated. Don't signal single step via SIGTRAP. | |
391 | */ | |
392 | clear_tsk_thread_flag(current, TIF_SINGLE_STEP); | |
393 | return; | |
394 | ||
395 | /* | |
396 | * Something tried to access memory that isn't in our memory map.. | |
397 | * Fix it, but check if it's kernel or user first.. | |
398 | */ | |
399 | bad_area: | |
400 | up_read(&mm->mmap_sem); | |
401 | ||
402 | /* User mode accesses just cause a SIGSEGV */ | |
403 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
404 | tsk->thread.prot_addr = address; | |
405 | tsk->thread.trap_no = error_code; | |
406 | do_sigsegv(regs, error_code, si_code, address); | |
407 | return; | |
408 | } | |
409 | ||
410 | no_context: | |
411 | /* Are we prepared to handle this kernel fault? */ | |
412 | fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK); | |
413 | if (fixup) { | |
414 | regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE; | |
415 | return; | |
416 | } | |
417 | ||
418 | /* | |
419 | * Oops. The kernel tried to access some bad page. We'll have to | |
420 | * terminate things with extreme prejudice. | |
421 | */ | |
422 | if (user_address == 0) | |
423 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" | |
424 | " at virtual kernel address %p\n", (void *)address); | |
425 | else | |
426 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
427 | " at virtual user address %p\n", (void *)address); | |
428 | ||
429 | die("Oops", regs, error_code); | |
430 | do_exit(SIGKILL); | |
431 | ||
432 | ||
433 | /* | |
434 | * We ran out of memory, or some other thing happened to us that made | |
435 | * us unable to handle the page fault gracefully. | |
436 | */ | |
437 | out_of_memory: | |
438 | up_read(&mm->mmap_sem); | |
f400e198 | 439 | if (is_init(tsk)) { |
1da177e4 | 440 | yield(); |
bac9c66c | 441 | down_read(&mm->mmap_sem); |
1da177e4 LT |
442 | goto survive; |
443 | } | |
444 | printk("VM: killing process %s\n", tsk->comm); | |
445 | if (regs->psw.mask & PSW_MASK_PSTATE) | |
446 | do_exit(SIGKILL); | |
447 | goto no_context; | |
448 | ||
449 | do_sigbus: | |
450 | up_read(&mm->mmap_sem); | |
451 | ||
452 | /* | |
453 | * Send a sigbus, regardless of whether we were in kernel | |
454 | * or user mode. | |
455 | */ | |
456 | tsk->thread.prot_addr = address; | |
457 | tsk->thread.trap_no = error_code; | |
458 | force_sig(SIGBUS, tsk); | |
459 | ||
460 | /* Kernel mode? Handle exceptions or die */ | |
461 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
462 | goto no_context; | |
463 | } | |
464 | ||
465 | void do_protection_exception(struct pt_regs *regs, unsigned long error_code) | |
466 | { | |
467 | regs->psw.addr -= (error_code >> 16); | |
468 | do_exception(regs, 4, 1); | |
469 | } | |
470 | ||
471 | void do_dat_exception(struct pt_regs *regs, unsigned long error_code) | |
472 | { | |
473 | do_exception(regs, error_code & 0xff, 0); | |
474 | } | |
475 | ||
1da177e4 LT |
476 | #ifdef CONFIG_PFAULT |
477 | /* | |
478 | * 'pfault' pseudo page faults routines. | |
479 | */ | |
29b08d2b | 480 | static ext_int_info_t ext_int_pfault; |
1da177e4 LT |
481 | static int pfault_disable = 0; |
482 | ||
483 | static int __init nopfault(char *str) | |
484 | { | |
485 | pfault_disable = 1; | |
486 | return 1; | |
487 | } | |
488 | ||
489 | __setup("nopfault", nopfault); | |
490 | ||
491 | typedef struct { | |
492 | __u16 refdiagc; | |
493 | __u16 reffcode; | |
494 | __u16 refdwlen; | |
495 | __u16 refversn; | |
496 | __u64 refgaddr; | |
497 | __u64 refselmk; | |
498 | __u64 refcmpmk; | |
499 | __u64 reserved; | |
500 | } __attribute__ ((packed)) pfault_refbk_t; | |
501 | ||
502 | int pfault_init(void) | |
503 | { | |
504 | pfault_refbk_t refbk = | |
505 | { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48, | |
506 | __PF_RES_FIELD }; | |
507 | int rc; | |
508 | ||
29b08d2b | 509 | if (!MACHINE_IS_VM || pfault_disable) |
1da177e4 | 510 | return -1; |
94c12cc7 MS |
511 | asm volatile( |
512 | " diag %1,%0,0x258\n" | |
513 | "0: j 2f\n" | |
514 | "1: la %0,8\n" | |
1da177e4 | 515 | "2:\n" |
94c12cc7 MS |
516 | EX_TABLE(0b,1b) |
517 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
518 | __ctl_set_bit(0, 9); |
519 | return rc; | |
520 | } | |
521 | ||
522 | void pfault_fini(void) | |
523 | { | |
524 | pfault_refbk_t refbk = | |
525 | { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL }; | |
526 | ||
29b08d2b | 527 | if (!MACHINE_IS_VM || pfault_disable) |
1da177e4 LT |
528 | return; |
529 | __ctl_clear_bit(0,9); | |
94c12cc7 MS |
530 | asm volatile( |
531 | " diag %0,0,0x258\n" | |
1da177e4 | 532 | "0:\n" |
94c12cc7 MS |
533 | EX_TABLE(0b,0b) |
534 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
535 | } |
536 | ||
2b67fc46 | 537 | static void pfault_interrupt(__u16 error_code) |
1da177e4 LT |
538 | { |
539 | struct task_struct *tsk; | |
540 | __u16 subcode; | |
541 | ||
542 | /* | |
543 | * Get the external interruption subcode & pfault | |
544 | * initial/completion signal bit. VM stores this | |
545 | * in the 'cpu address' field associated with the | |
546 | * external interrupt. | |
547 | */ | |
548 | subcode = S390_lowcore.cpu_addr; | |
549 | if ((subcode & 0xff00) != __SUBCODE_MASK) | |
550 | return; | |
551 | ||
552 | /* | |
553 | * Get the token (= address of the task structure of the affected task). | |
554 | */ | |
555 | tsk = *(struct task_struct **) __LC_PFAULT_INTPARM; | |
556 | ||
557 | if (subcode & 0x0080) { | |
558 | /* signal bit is set -> a page has been swapped in by VM */ | |
559 | if (xchg(&tsk->thread.pfault_wait, -1) != 0) { | |
560 | /* Initial interrupt was faster than the completion | |
561 | * interrupt. pfault_wait is valid. Set pfault_wait | |
562 | * back to zero and wake up the process. This can | |
563 | * safely be done because the task is still sleeping | |
b6d09449 | 564 | * and can't produce new pfaults. */ |
1da177e4 LT |
565 | tsk->thread.pfault_wait = 0; |
566 | wake_up_process(tsk); | |
b6d09449 | 567 | put_task_struct(tsk); |
1da177e4 LT |
568 | } |
569 | } else { | |
570 | /* signal bit not set -> a real page is missing. */ | |
b6d09449 | 571 | get_task_struct(tsk); |
1da177e4 LT |
572 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
573 | if (xchg(&tsk->thread.pfault_wait, 1) != 0) { | |
574 | /* Completion interrupt was faster than the initial | |
575 | * interrupt (swapped in a -1 for pfault_wait). Set | |
576 | * pfault_wait back to zero and exit. This can be | |
577 | * done safely because tsk is running in kernel | |
578 | * mode and can't produce new pfaults. */ | |
579 | tsk->thread.pfault_wait = 0; | |
580 | set_task_state(tsk, TASK_RUNNING); | |
b6d09449 | 581 | put_task_struct(tsk); |
1da177e4 LT |
582 | } else |
583 | set_tsk_need_resched(tsk); | |
584 | } | |
585 | } | |
1da177e4 | 586 | |
29b08d2b HC |
587 | void __init pfault_irq_init(void) |
588 | { | |
589 | if (!MACHINE_IS_VM) | |
590 | return; | |
591 | ||
592 | /* | |
593 | * Try to get pfault pseudo page faults going. | |
594 | */ | |
595 | if (register_early_external_interrupt(0x2603, pfault_interrupt, | |
596 | &ext_int_pfault) != 0) | |
597 | panic("Couldn't request external interrupt 0x2603"); | |
598 | ||
599 | if (pfault_init() == 0) | |
600 | return; | |
601 | ||
602 | /* Tough luck, no pfault. */ | |
603 | pfault_disable = 1; | |
604 | unregister_early_external_interrupt(0x2603, pfault_interrupt, | |
605 | &ext_int_pfault); | |
606 | } | |
607 | #endif |