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 | ||
052ff461 | 13 | #include <linux/kernel_stat.h> |
cdd6c482 | 14 | #include <linux/perf_event.h> |
1da177e4 LT |
15 | #include <linux/signal.h> |
16 | #include <linux/sched.h> | |
17 | #include <linux/kernel.h> | |
18 | #include <linux/errno.h> | |
19 | #include <linux/string.h> | |
20 | #include <linux/types.h> | |
21 | #include <linux/ptrace.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/mm.h> | |
7757591a | 24 | #include <linux/compat.h> |
1da177e4 | 25 | #include <linux/smp.h> |
1eeb66a1 | 26 | #include <linux/kdebug.h> |
1da177e4 LT |
27 | #include <linux/init.h> |
28 | #include <linux/console.h> | |
29 | #include <linux/module.h> | |
30 | #include <linux/hardirq.h> | |
4ba069b8 | 31 | #include <linux/kprobes.h> |
be5ec363 | 32 | #include <linux/uaccess.h> |
53492b1d | 33 | #include <linux/hugetlb.h> |
cbb870c8 | 34 | #include <asm/asm-offsets.h> |
1da177e4 | 35 | #include <asm/system.h> |
1da177e4 | 36 | #include <asm/pgtable.h> |
d7b250e2 | 37 | #include <asm/irq.h> |
6252d702 | 38 | #include <asm/mmu_context.h> |
50d7280d | 39 | #include <asm/compat.h> |
a806170e | 40 | #include "../kernel/entry.h" |
1da177e4 | 41 | |
347a8dc3 | 42 | #ifndef CONFIG_64BIT |
1da177e4 | 43 | #define __FAIL_ADDR_MASK 0x7ffff000 |
1da177e4 LT |
44 | #define __SUBCODE_MASK 0x0200 |
45 | #define __PF_RES_FIELD 0ULL | |
347a8dc3 | 46 | #else /* CONFIG_64BIT */ |
1da177e4 | 47 | #define __FAIL_ADDR_MASK -4096L |
1da177e4 LT |
48 | #define __SUBCODE_MASK 0x0600 |
49 | #define __PF_RES_FIELD 0x8000000000000000ULL | |
347a8dc3 | 50 | #endif /* CONFIG_64BIT */ |
1da177e4 | 51 | |
50d7280d MS |
52 | #define VM_FAULT_BADCONTEXT 0x010000 |
53 | #define VM_FAULT_BADMAP 0x020000 | |
54 | #define VM_FAULT_BADACCESS 0x040000 | |
55 | ||
92f842ea MS |
56 | static unsigned long store_indication; |
57 | ||
58 | void fault_init(void) | |
59 | { | |
14375bc4 | 60 | if (test_facility(2) && test_facility(75)) |
92f842ea MS |
61 | store_indication = 0xc00; |
62 | } | |
63 | ||
7ecb344a | 64 | static inline int notify_page_fault(struct pt_regs *regs) |
10c1031f | 65 | { |
33464e3b CH |
66 | int ret = 0; |
67 | ||
68 | /* kprobe_running() needs smp_processor_id() */ | |
22e0a046 | 69 | if (kprobes_built_in() && !user_mode(regs)) { |
33464e3b CH |
70 | preempt_disable(); |
71 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
72 | ret = 1; | |
73 | preempt_enable(); | |
74 | } | |
33464e3b | 75 | return ret; |
4ba069b8 | 76 | } |
4ba069b8 | 77 | |
1da177e4 LT |
78 | |
79 | /* | |
80 | * Unlock any spinlocks which will prevent us from getting the | |
cefc8be8 | 81 | * message out. |
1da177e4 LT |
82 | */ |
83 | void bust_spinlocks(int yes) | |
84 | { | |
85 | if (yes) { | |
86 | oops_in_progress = 1; | |
87 | } else { | |
88 | int loglevel_save = console_loglevel; | |
89 | console_unblank(); | |
90 | oops_in_progress = 0; | |
91 | /* | |
92 | * OK, the message is on the console. Now we call printk() | |
93 | * without oops_in_progress set so that printk will give klogd | |
94 | * a poke. Hold onto your hats... | |
95 | */ | |
96 | console_loglevel = 15; | |
97 | printk(" "); | |
98 | console_loglevel = loglevel_save; | |
99 | } | |
100 | } | |
101 | ||
102 | /* | |
482b05dd | 103 | * Returns the address space associated with the fault. |
61365e13 | 104 | * Returns 0 for kernel space and 1 for user space. |
1da177e4 | 105 | */ |
61365e13 | 106 | static inline int user_space_fault(unsigned long trans_exc_code) |
1da177e4 LT |
107 | { |
108 | /* | |
61365e13 MS |
109 | * The lowest two bits of the translation exception |
110 | * identification indicate which paging table was used. | |
1da177e4 | 111 | */ |
61365e13 MS |
112 | trans_exc_code &= 3; |
113 | if (trans_exc_code == 2) | |
114 | /* Access via secondary space, set_fs setting decides */ | |
115 | return current->thread.mm_segment.ar4; | |
b11b5334 | 116 | if (user_mode == HOME_SPACE_MODE) |
61365e13 MS |
117 | /* User space if the access has been done via home space. */ |
118 | return trans_exc_code == 3; | |
119 | /* | |
120 | * If the user space is not the home space the kernel runs in home | |
121 | * space. Access via secondary space has already been covered, | |
122 | * access via primary space or access register is from user space | |
123 | * and access via home space is from the kernel. | |
124 | */ | |
125 | return trans_exc_code != 3; | |
1da177e4 LT |
126 | } |
127 | ||
ab3c68ee HC |
128 | static inline void report_user_fault(struct pt_regs *regs, long int_code, |
129 | int signr, unsigned long address) | |
130 | { | |
131 | if ((task_pid_nr(current) > 1) && !show_unhandled_signals) | |
132 | return; | |
133 | if (!unhandled_signal(current, signr)) | |
134 | return; | |
135 | if (!printk_ratelimit()) | |
136 | return; | |
137 | printk("User process fault: interruption code 0x%lX ", int_code); | |
138 | print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN); | |
139 | printk("\n"); | |
140 | printk("failing address: %lX\n", address); | |
141 | show_regs(regs); | |
142 | } | |
143 | ||
1da177e4 LT |
144 | /* |
145 | * Send SIGSEGV to task. This is an external routine | |
146 | * to keep the stack usage of do_page_fault small. | |
147 | */ | |
50d7280d MS |
148 | static noinline void do_sigsegv(struct pt_regs *regs, long int_code, |
149 | int si_code, unsigned long trans_exc_code) | |
1da177e4 LT |
150 | { |
151 | struct siginfo si; | |
50d7280d | 152 | unsigned long address; |
1da177e4 | 153 | |
50d7280d MS |
154 | address = trans_exc_code & __FAIL_ADDR_MASK; |
155 | current->thread.prot_addr = address; | |
156 | current->thread.trap_no = int_code; | |
ab3c68ee | 157 | report_user_fault(regs, int_code, SIGSEGV, address); |
1da177e4 LT |
158 | si.si_signo = SIGSEGV; |
159 | si.si_code = si_code; | |
d2c993d8 | 160 | si.si_addr = (void __user *) address; |
1da177e4 LT |
161 | force_sig_info(SIGSEGV, &si, current); |
162 | } | |
163 | ||
50d7280d MS |
164 | static noinline void do_no_context(struct pt_regs *regs, long int_code, |
165 | unsigned long trans_exc_code) | |
10c1031f MS |
166 | { |
167 | const struct exception_table_entry *fixup; | |
61365e13 | 168 | unsigned long address; |
10c1031f MS |
169 | |
170 | /* Are we prepared to handle this kernel fault? */ | |
50d7280d | 171 | fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN); |
10c1031f MS |
172 | if (fixup) { |
173 | regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE; | |
174 | return; | |
175 | } | |
176 | ||
177 | /* | |
178 | * Oops. The kernel tried to access some bad page. We'll have to | |
179 | * terminate things with extreme prejudice. | |
180 | */ | |
61365e13 | 181 | address = trans_exc_code & __FAIL_ADDR_MASK; |
b11b5334 | 182 | if (!user_space_fault(trans_exc_code)) |
10c1031f MS |
183 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" |
184 | " at virtual kernel address %p\n", (void *)address); | |
185 | else | |
186 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
187 | " at virtual user address %p\n", (void *)address); | |
188 | ||
50d7280d | 189 | die("Oops", regs, int_code); |
10c1031f MS |
190 | do_exit(SIGKILL); |
191 | } | |
192 | ||
50d7280d MS |
193 | static noinline void do_low_address(struct pt_regs *regs, long int_code, |
194 | unsigned long trans_exc_code) | |
10c1031f MS |
195 | { |
196 | /* Low-address protection hit in kernel mode means | |
197 | NULL pointer write access in kernel mode. */ | |
198 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
199 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
50d7280d | 200 | die ("Low-address protection", regs, int_code); |
10c1031f MS |
201 | do_exit(SIGKILL); |
202 | } | |
203 | ||
50d7280d | 204 | do_no_context(regs, int_code, trans_exc_code); |
10c1031f MS |
205 | } |
206 | ||
50d7280d MS |
207 | static noinline void do_sigbus(struct pt_regs *regs, long int_code, |
208 | unsigned long trans_exc_code) | |
10c1031f MS |
209 | { |
210 | struct task_struct *tsk = current; | |
36bf9680 MS |
211 | unsigned long address; |
212 | struct siginfo si; | |
10c1031f | 213 | |
10c1031f MS |
214 | /* |
215 | * Send a sigbus, regardless of whether we were in kernel | |
216 | * or user mode. | |
217 | */ | |
36bf9680 MS |
218 | address = trans_exc_code & __FAIL_ADDR_MASK; |
219 | tsk->thread.prot_addr = address; | |
50d7280d | 220 | tsk->thread.trap_no = int_code; |
36bf9680 MS |
221 | si.si_signo = SIGBUS; |
222 | si.si_errno = 0; | |
223 | si.si_code = BUS_ADRERR; | |
224 | si.si_addr = (void __user *) address; | |
225 | force_sig_info(SIGBUS, &si, tsk); | |
10c1031f MS |
226 | } |
227 | ||
50d7280d MS |
228 | static noinline void do_fault_error(struct pt_regs *regs, long int_code, |
229 | unsigned long trans_exc_code, int fault) | |
230 | { | |
231 | int si_code; | |
232 | ||
233 | switch (fault) { | |
234 | case VM_FAULT_BADACCESS: | |
50d7280d MS |
235 | case VM_FAULT_BADMAP: |
236 | /* Bad memory access. Check if it is kernel or user space. */ | |
237 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
238 | /* User mode accesses just cause a SIGSEGV */ | |
239 | si_code = (fault == VM_FAULT_BADMAP) ? | |
240 | SEGV_MAPERR : SEGV_ACCERR; | |
241 | do_sigsegv(regs, int_code, si_code, trans_exc_code); | |
242 | return; | |
243 | } | |
244 | case VM_FAULT_BADCONTEXT: | |
245 | do_no_context(regs, int_code, trans_exc_code); | |
246 | break; | |
247 | default: /* fault & VM_FAULT_ERROR */ | |
99583181 HC |
248 | if (fault & VM_FAULT_OOM) { |
249 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
250 | do_no_context(regs, int_code, trans_exc_code); | |
251 | else | |
252 | pagefault_out_of_memory(); | |
253 | } else if (fault & VM_FAULT_SIGBUS) { | |
50d7280d MS |
254 | /* Kernel mode? Handle exceptions or die */ |
255 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
256 | do_no_context(regs, int_code, trans_exc_code); | |
36bf9680 MS |
257 | else |
258 | do_sigbus(regs, int_code, trans_exc_code); | |
50d7280d MS |
259 | } else |
260 | BUG(); | |
261 | break; | |
262 | } | |
263 | } | |
264 | ||
1da177e4 LT |
265 | /* |
266 | * This routine handles page faults. It determines the address, | |
267 | * and the problem, and then passes it off to one of the appropriate | |
268 | * routines. | |
269 | * | |
50d7280d | 270 | * interruption code (int_code): |
1da177e4 LT |
271 | * 04 Protection -> Write-Protection (suprression) |
272 | * 10 Segment translation -> Not present (nullification) | |
273 | * 11 Page translation -> Not present (nullification) | |
274 | * 3b Region third trans. -> Not present (nullification) | |
275 | */ | |
1ab947de | 276 | static inline int do_exception(struct pt_regs *regs, int access, |
50d7280d | 277 | unsigned long trans_exc_code) |
1da177e4 | 278 | { |
10c1031f MS |
279 | struct task_struct *tsk; |
280 | struct mm_struct *mm; | |
281 | struct vm_area_struct *vma; | |
282 | unsigned long address; | |
92f842ea | 283 | int fault, write; |
1da177e4 | 284 | |
7ecb344a | 285 | if (notify_page_fault(regs)) |
50d7280d | 286 | return 0; |
4ba069b8 | 287 | |
10c1031f MS |
288 | tsk = current; |
289 | mm = tsk->mm; | |
1da177e4 | 290 | |
1da177e4 LT |
291 | /* |
292 | * Verify that the fault happened in user space, that | |
293 | * we are not in an interrupt and that there is a | |
294 | * user context. | |
295 | */ | |
50d7280d | 296 | fault = VM_FAULT_BADCONTEXT; |
61365e13 | 297 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
50d7280d | 298 | goto out; |
1da177e4 | 299 | |
61365e13 | 300 | address = trans_exc_code & __FAIL_ADDR_MASK; |
cdd6c482 | 301 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); |
10c1031f | 302 | down_read(&mm->mmap_sem); |
1da177e4 | 303 | |
50d7280d | 304 | fault = VM_FAULT_BADMAP; |
482b05dd GS |
305 | vma = find_vma(mm, address); |
306 | if (!vma) | |
50d7280d | 307 | goto out_up; |
c1821c2e | 308 | |
50d7280d MS |
309 | if (unlikely(vma->vm_start > address)) { |
310 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
311 | goto out_up; | |
312 | if (expand_stack(vma, address)) | |
313 | goto out_up; | |
314 | } | |
315 | ||
316 | /* | |
317 | * Ok, we have a good vm_area for this memory access, so | |
318 | * we can handle it.. | |
319 | */ | |
320 | fault = VM_FAULT_BADACCESS; | |
1ab947de | 321 | if (unlikely(!(vma->vm_flags & access))) |
50d7280d | 322 | goto out_up; |
1da177e4 | 323 | |
53492b1d GS |
324 | if (is_vm_hugetlb_page(vma)) |
325 | address &= HPAGE_MASK; | |
1da177e4 LT |
326 | /* |
327 | * If for any reason at all we couldn't handle the fault, | |
328 | * make sure we exit gracefully rather than endlessly redo | |
329 | * the fault. | |
330 | */ | |
92f842ea MS |
331 | write = (access == VM_WRITE || |
332 | (trans_exc_code & store_indication) == 0x400) ? | |
333 | FAULT_FLAG_WRITE : 0; | |
334 | fault = handle_mm_fault(mm, vma, address, write); | |
50d7280d MS |
335 | if (unlikely(fault & VM_FAULT_ERROR)) |
336 | goto out_up; | |
337 | ||
bde69af2 | 338 | if (fault & VM_FAULT_MAJOR) { |
83c54070 | 339 | tsk->maj_flt++; |
cdd6c482 | 340 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, |
bde69af2 HC |
341 | regs, address); |
342 | } else { | |
83c54070 | 343 | tsk->min_flt++; |
cdd6c482 | 344 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, |
bde69af2 HC |
345 | regs, address); |
346 | } | |
1da177e4 LT |
347 | /* |
348 | * The instruction that caused the program check will | |
349 | * be repeated. Don't signal single step via SIGTRAP. | |
350 | */ | |
5e9a2692 | 351 | clear_tsk_thread_flag(tsk, TIF_PER_TRAP); |
50d7280d MS |
352 | fault = 0; |
353 | out_up: | |
10c1031f | 354 | up_read(&mm->mmap_sem); |
50d7280d MS |
355 | out: |
356 | return fault; | |
1da177e4 LT |
357 | } |
358 | ||
1e54622e MS |
359 | void __kprobes do_protection_exception(struct pt_regs *regs, long pgm_int_code, |
360 | unsigned long trans_exc_code) | |
1da177e4 | 361 | { |
50d7280d | 362 | int fault; |
61365e13 | 363 | |
25985edc | 364 | /* Protection exception is suppressing, decrement psw address. */ |
1e54622e | 365 | regs->psw.addr -= (pgm_int_code >> 16); |
10c1031f MS |
366 | /* |
367 | * Check for low-address protection. This needs to be treated | |
368 | * as a special case because the translation exception code | |
369 | * field is not guaranteed to contain valid data in this case. | |
370 | */ | |
61365e13 | 371 | if (unlikely(!(trans_exc_code & 4))) { |
1e54622e | 372 | do_low_address(regs, pgm_int_code, trans_exc_code); |
10c1031f MS |
373 | return; |
374 | } | |
1ab947de | 375 | fault = do_exception(regs, VM_WRITE, trans_exc_code); |
50d7280d MS |
376 | if (unlikely(fault)) |
377 | do_fault_error(regs, 4, trans_exc_code, fault); | |
1da177e4 LT |
378 | } |
379 | ||
1e54622e MS |
380 | void __kprobes do_dat_exception(struct pt_regs *regs, long pgm_int_code, |
381 | unsigned long trans_exc_code) | |
1da177e4 | 382 | { |
1ab947de | 383 | int access, fault; |
50d7280d | 384 | |
1ab947de | 385 | access = VM_READ | VM_EXEC | VM_WRITE; |
1ab947de | 386 | fault = do_exception(regs, access, trans_exc_code); |
50d7280d | 387 | if (unlikely(fault)) |
1e54622e | 388 | do_fault_error(regs, pgm_int_code & 255, trans_exc_code, fault); |
1da177e4 LT |
389 | } |
390 | ||
6252d702 | 391 | #ifdef CONFIG_64BIT |
1e54622e MS |
392 | void __kprobes do_asce_exception(struct pt_regs *regs, long pgm_int_code, |
393 | unsigned long trans_exc_code) | |
6252d702 | 394 | { |
50d7280d | 395 | struct mm_struct *mm = current->mm; |
6252d702 | 396 | struct vm_area_struct *vma; |
6252d702 | 397 | |
61365e13 | 398 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
6252d702 MS |
399 | goto no_context; |
400 | ||
6252d702 | 401 | down_read(&mm->mmap_sem); |
50d7280d | 402 | vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK); |
6252d702 MS |
403 | up_read(&mm->mmap_sem); |
404 | ||
405 | if (vma) { | |
406 | update_mm(mm, current); | |
407 | return; | |
408 | } | |
409 | ||
410 | /* User mode accesses just cause a SIGSEGV */ | |
411 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
1e54622e | 412 | do_sigsegv(regs, pgm_int_code, SEGV_MAPERR, trans_exc_code); |
6252d702 MS |
413 | return; |
414 | } | |
415 | ||
416 | no_context: | |
1e54622e | 417 | do_no_context(regs, pgm_int_code, trans_exc_code); |
6252d702 MS |
418 | } |
419 | #endif | |
420 | ||
1e54622e | 421 | int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write) |
6c1e3e79 GS |
422 | { |
423 | struct pt_regs regs; | |
424 | int access, fault; | |
425 | ||
426 | regs.psw.mask = psw_kernel_bits; | |
427 | if (!irqs_disabled()) | |
428 | regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT; | |
429 | regs.psw.addr = (unsigned long) __builtin_return_address(0); | |
430 | regs.psw.addr |= PSW_ADDR_AMODE; | |
431 | uaddr &= PAGE_MASK; | |
1e54622e | 432 | access = write ? VM_WRITE : VM_READ; |
6c1e3e79 GS |
433 | fault = do_exception(®s, access, uaddr | 2); |
434 | if (unlikely(fault)) { | |
99583181 HC |
435 | if (fault & VM_FAULT_OOM) |
436 | return -EFAULT; | |
437 | else if (fault & VM_FAULT_SIGBUS) | |
1e54622e | 438 | do_sigbus(®s, pgm_int_code, uaddr); |
6c1e3e79 GS |
439 | } |
440 | return fault ? -EFAULT : 0; | |
441 | } | |
442 | ||
1da177e4 LT |
443 | #ifdef CONFIG_PFAULT |
444 | /* | |
445 | * 'pfault' pseudo page faults routines. | |
446 | */ | |
fb0a9d7e | 447 | static int pfault_disable; |
1da177e4 LT |
448 | |
449 | static int __init nopfault(char *str) | |
450 | { | |
451 | pfault_disable = 1; | |
452 | return 1; | |
453 | } | |
454 | ||
455 | __setup("nopfault", nopfault); | |
456 | ||
7dd8fe1f HC |
457 | struct pfault_refbk { |
458 | u16 refdiagc; | |
459 | u16 reffcode; | |
460 | u16 refdwlen; | |
461 | u16 refversn; | |
462 | u64 refgaddr; | |
463 | u64 refselmk; | |
464 | u64 refcmpmk; | |
465 | u64 reserved; | |
466 | } __attribute__ ((packed, aligned(8))); | |
1da177e4 LT |
467 | |
468 | int pfault_init(void) | |
469 | { | |
7dd8fe1f HC |
470 | struct pfault_refbk refbk = { |
471 | .refdiagc = 0x258, | |
472 | .reffcode = 0, | |
473 | .refdwlen = 5, | |
474 | .refversn = 2, | |
475 | .refgaddr = __LC_CURRENT_PID, | |
476 | .refselmk = 1ULL << 48, | |
477 | .refcmpmk = 1ULL << 48, | |
478 | .reserved = __PF_RES_FIELD }; | |
1da177e4 LT |
479 | int rc; |
480 | ||
29b08d2b | 481 | if (!MACHINE_IS_VM || pfault_disable) |
1da177e4 | 482 | return -1; |
94c12cc7 MS |
483 | asm volatile( |
484 | " diag %1,%0,0x258\n" | |
485 | "0: j 2f\n" | |
486 | "1: la %0,8\n" | |
1da177e4 | 487 | "2:\n" |
94c12cc7 MS |
488 | EX_TABLE(0b,1b) |
489 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
490 | return rc; |
491 | } | |
492 | ||
493 | void pfault_fini(void) | |
494 | { | |
7dd8fe1f HC |
495 | struct pfault_refbk refbk = { |
496 | .refdiagc = 0x258, | |
497 | .reffcode = 1, | |
498 | .refdwlen = 5, | |
499 | .refversn = 2, | |
500 | }; | |
1da177e4 | 501 | |
29b08d2b | 502 | if (!MACHINE_IS_VM || pfault_disable) |
1da177e4 | 503 | return; |
94c12cc7 MS |
504 | asm volatile( |
505 | " diag %0,0,0x258\n" | |
1da177e4 | 506 | "0:\n" |
94c12cc7 MS |
507 | EX_TABLE(0b,0b) |
508 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
509 | } |
510 | ||
f2db2e6c HC |
511 | static DEFINE_SPINLOCK(pfault_lock); |
512 | static LIST_HEAD(pfault_list); | |
513 | ||
f6649a7e MS |
514 | static void pfault_interrupt(unsigned int ext_int_code, |
515 | unsigned int param32, unsigned long param64) | |
1da177e4 LT |
516 | { |
517 | struct task_struct *tsk; | |
518 | __u16 subcode; | |
f2db2e6c | 519 | pid_t pid; |
1da177e4 LT |
520 | |
521 | /* | |
522 | * Get the external interruption subcode & pfault | |
523 | * initial/completion signal bit. VM stores this | |
524 | * in the 'cpu address' field associated with the | |
525 | * external interrupt. | |
526 | */ | |
f6649a7e | 527 | subcode = ext_int_code >> 16; |
1da177e4 LT |
528 | if ((subcode & 0xff00) != __SUBCODE_MASK) |
529 | return; | |
a9851832 | 530 | kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++; |
f2db2e6c HC |
531 | if (subcode & 0x0080) { |
532 | /* Get the token (= pid of the affected task). */ | |
533 | pid = sizeof(void *) == 4 ? param32 : param64; | |
534 | rcu_read_lock(); | |
535 | tsk = find_task_by_pid_ns(pid, &init_pid_ns); | |
536 | if (tsk) | |
537 | get_task_struct(tsk); | |
538 | rcu_read_unlock(); | |
539 | if (!tsk) | |
540 | return; | |
541 | } else { | |
542 | tsk = current; | |
543 | } | |
544 | spin_lock(&pfault_lock); | |
1da177e4 LT |
545 | if (subcode & 0x0080) { |
546 | /* signal bit is set -> a page has been swapped in by VM */ | |
f2db2e6c | 547 | if (tsk->thread.pfault_wait == 1) { |
1da177e4 LT |
548 | /* Initial interrupt was faster than the completion |
549 | * interrupt. pfault_wait is valid. Set pfault_wait | |
550 | * back to zero and wake up the process. This can | |
551 | * safely be done because the task is still sleeping | |
b6d09449 | 552 | * and can't produce new pfaults. */ |
1da177e4 | 553 | tsk->thread.pfault_wait = 0; |
f2db2e6c | 554 | list_del(&tsk->thread.list); |
1da177e4 | 555 | wake_up_process(tsk); |
f2db2e6c HC |
556 | } else { |
557 | /* Completion interrupt was faster than initial | |
558 | * interrupt. Set pfault_wait to -1 so the initial | |
559 | * interrupt doesn't put the task to sleep. */ | |
560 | tsk->thread.pfault_wait = -1; | |
1da177e4 | 561 | } |
f2db2e6c | 562 | put_task_struct(tsk); |
1da177e4 LT |
563 | } else { |
564 | /* signal bit not set -> a real page is missing. */ | |
f2db2e6c | 565 | if (tsk->thread.pfault_wait == -1) { |
1da177e4 | 566 | /* Completion interrupt was faster than the initial |
f2db2e6c HC |
567 | * interrupt (pfault_wait == -1). Set pfault_wait |
568 | * back to zero and exit. */ | |
1da177e4 | 569 | tsk->thread.pfault_wait = 0; |
f2db2e6c HC |
570 | } else { |
571 | /* Initial interrupt arrived before completion | |
572 | * interrupt. Let the task sleep. */ | |
573 | tsk->thread.pfault_wait = 1; | |
574 | list_add(&tsk->thread.list, &pfault_list); | |
575 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 576 | set_tsk_need_resched(tsk); |
f2db2e6c HC |
577 | } |
578 | } | |
579 | spin_unlock(&pfault_lock); | |
580 | } | |
581 | ||
582 | static int __cpuinit pfault_cpu_notify(struct notifier_block *self, | |
583 | unsigned long action, void *hcpu) | |
584 | { | |
585 | struct thread_struct *thread, *next; | |
586 | struct task_struct *tsk; | |
587 | ||
588 | switch (action) { | |
589 | case CPU_DEAD: | |
590 | case CPU_DEAD_FROZEN: | |
591 | spin_lock_irq(&pfault_lock); | |
592 | list_for_each_entry_safe(thread, next, &pfault_list, list) { | |
593 | thread->pfault_wait = 0; | |
594 | list_del(&thread->list); | |
595 | tsk = container_of(thread, struct task_struct, thread); | |
596 | wake_up_process(tsk); | |
597 | } | |
598 | spin_unlock_irq(&pfault_lock); | |
599 | break; | |
600 | default: | |
601 | break; | |
1da177e4 | 602 | } |
f2db2e6c | 603 | return NOTIFY_OK; |
1da177e4 | 604 | } |
1da177e4 | 605 | |
fb0a9d7e | 606 | static int __init pfault_irq_init(void) |
29b08d2b | 607 | { |
fb0a9d7e | 608 | int rc; |
29b08d2b | 609 | |
fb0a9d7e HC |
610 | if (!MACHINE_IS_VM) |
611 | return 0; | |
fb0a9d7e | 612 | rc = register_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
613 | if (rc) |
614 | goto out_extint; | |
615 | rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; | |
616 | if (rc) | |
617 | goto out_pfault; | |
df7997ab | 618 | service_subclass_irq_register(); |
7dd8fe1f HC |
619 | hotcpu_notifier(pfault_cpu_notify, 0); |
620 | return 0; | |
29b08d2b | 621 | |
7dd8fe1f | 622 | out_pfault: |
fb0a9d7e | 623 | unregister_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
624 | out_extint: |
625 | pfault_disable = 1; | |
626 | return rc; | |
29b08d2b | 627 | } |
fb0a9d7e HC |
628 | early_initcall(pfault_irq_init); |
629 | ||
7dd8fe1f | 630 | #endif /* CONFIG_PFAULT */ |