]>
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 | ||
cdd6c482 | 13 | #include <linux/perf_event.h> |
1da177e4 LT |
14 | #include <linux/signal.h> |
15 | #include <linux/sched.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/errno.h> | |
18 | #include <linux/string.h> | |
19 | #include <linux/types.h> | |
20 | #include <linux/ptrace.h> | |
21 | #include <linux/mman.h> | |
22 | #include <linux/mm.h> | |
7757591a | 23 | #include <linux/compat.h> |
1da177e4 | 24 | #include <linux/smp.h> |
1eeb66a1 | 25 | #include <linux/kdebug.h> |
1da177e4 LT |
26 | #include <linux/init.h> |
27 | #include <linux/console.h> | |
28 | #include <linux/module.h> | |
29 | #include <linux/hardirq.h> | |
4ba069b8 | 30 | #include <linux/kprobes.h> |
be5ec363 | 31 | #include <linux/uaccess.h> |
53492b1d | 32 | #include <linux/hugetlb.h> |
cbb870c8 | 33 | #include <asm/asm-offsets.h> |
1da177e4 | 34 | #include <asm/system.h> |
1da177e4 | 35 | #include <asm/pgtable.h> |
29b08d2b | 36 | #include <asm/s390_ext.h> |
6252d702 | 37 | #include <asm/mmu_context.h> |
50d7280d | 38 | #include <asm/compat.h> |
a806170e | 39 | #include "../kernel/entry.h" |
1da177e4 | 40 | |
347a8dc3 | 41 | #ifndef CONFIG_64BIT |
1da177e4 | 42 | #define __FAIL_ADDR_MASK 0x7ffff000 |
1da177e4 LT |
43 | #define __SUBCODE_MASK 0x0200 |
44 | #define __PF_RES_FIELD 0ULL | |
347a8dc3 | 45 | #else /* CONFIG_64BIT */ |
1da177e4 | 46 | #define __FAIL_ADDR_MASK -4096L |
1da177e4 LT |
47 | #define __SUBCODE_MASK 0x0600 |
48 | #define __PF_RES_FIELD 0x8000000000000000ULL | |
347a8dc3 | 49 | #endif /* CONFIG_64BIT */ |
1da177e4 | 50 | |
50d7280d MS |
51 | #define VM_FAULT_BADCONTEXT 0x010000 |
52 | #define VM_FAULT_BADMAP 0x020000 | |
53 | #define VM_FAULT_BADACCESS 0x040000 | |
54 | ||
92f842ea MS |
55 | static unsigned long store_indication; |
56 | ||
57 | void fault_init(void) | |
58 | { | |
59 | unsigned long long facility_list[2]; | |
60 | ||
61 | if (stfle(facility_list, 2) < 2) | |
62 | return; | |
63 | if ((facility_list[0] & (1ULL << 61)) && | |
64 | (facility_list[1] & (1ULL << 52))) | |
65 | store_indication = 0xc00; | |
66 | } | |
67 | ||
7ecb344a | 68 | static inline int notify_page_fault(struct pt_regs *regs) |
10c1031f | 69 | { |
33464e3b CH |
70 | int ret = 0; |
71 | ||
72 | /* kprobe_running() needs smp_processor_id() */ | |
22e0a046 | 73 | if (kprobes_built_in() && !user_mode(regs)) { |
33464e3b CH |
74 | preempt_disable(); |
75 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
76 | ret = 1; | |
77 | preempt_enable(); | |
78 | } | |
33464e3b | 79 | return ret; |
4ba069b8 | 80 | } |
4ba069b8 | 81 | |
1da177e4 LT |
82 | |
83 | /* | |
84 | * Unlock any spinlocks which will prevent us from getting the | |
cefc8be8 | 85 | * message out. |
1da177e4 LT |
86 | */ |
87 | void bust_spinlocks(int yes) | |
88 | { | |
89 | if (yes) { | |
90 | oops_in_progress = 1; | |
91 | } else { | |
92 | int loglevel_save = console_loglevel; | |
93 | console_unblank(); | |
94 | oops_in_progress = 0; | |
95 | /* | |
96 | * OK, the message is on the console. Now we call printk() | |
97 | * without oops_in_progress set so that printk will give klogd | |
98 | * a poke. Hold onto your hats... | |
99 | */ | |
100 | console_loglevel = 15; | |
101 | printk(" "); | |
102 | console_loglevel = loglevel_save; | |
103 | } | |
104 | } | |
105 | ||
106 | /* | |
482b05dd | 107 | * Returns the address space associated with the fault. |
61365e13 | 108 | * Returns 0 for kernel space and 1 for user space. |
1da177e4 | 109 | */ |
61365e13 | 110 | static inline int user_space_fault(unsigned long trans_exc_code) |
1da177e4 LT |
111 | { |
112 | /* | |
61365e13 MS |
113 | * The lowest two bits of the translation exception |
114 | * identification indicate which paging table was used. | |
1da177e4 | 115 | */ |
61365e13 MS |
116 | trans_exc_code &= 3; |
117 | if (trans_exc_code == 2) | |
118 | /* Access via secondary space, set_fs setting decides */ | |
119 | return current->thread.mm_segment.ar4; | |
b11b5334 | 120 | if (user_mode == HOME_SPACE_MODE) |
61365e13 MS |
121 | /* User space if the access has been done via home space. */ |
122 | return trans_exc_code == 3; | |
123 | /* | |
124 | * If the user space is not the home space the kernel runs in home | |
125 | * space. Access via secondary space has already been covered, | |
126 | * access via primary space or access register is from user space | |
127 | * and access via home space is from the kernel. | |
128 | */ | |
129 | return trans_exc_code != 3; | |
1da177e4 LT |
130 | } |
131 | ||
ab3c68ee HC |
132 | static inline void report_user_fault(struct pt_regs *regs, long int_code, |
133 | int signr, unsigned long address) | |
134 | { | |
135 | if ((task_pid_nr(current) > 1) && !show_unhandled_signals) | |
136 | return; | |
137 | if (!unhandled_signal(current, signr)) | |
138 | return; | |
139 | if (!printk_ratelimit()) | |
140 | return; | |
141 | printk("User process fault: interruption code 0x%lX ", int_code); | |
142 | print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN); | |
143 | printk("\n"); | |
144 | printk("failing address: %lX\n", address); | |
145 | show_regs(regs); | |
146 | } | |
147 | ||
1da177e4 LT |
148 | /* |
149 | * Send SIGSEGV to task. This is an external routine | |
150 | * to keep the stack usage of do_page_fault small. | |
151 | */ | |
50d7280d MS |
152 | static noinline void do_sigsegv(struct pt_regs *regs, long int_code, |
153 | int si_code, unsigned long trans_exc_code) | |
1da177e4 LT |
154 | { |
155 | struct siginfo si; | |
50d7280d | 156 | unsigned long address; |
1da177e4 | 157 | |
50d7280d MS |
158 | address = trans_exc_code & __FAIL_ADDR_MASK; |
159 | current->thread.prot_addr = address; | |
160 | current->thread.trap_no = int_code; | |
ab3c68ee | 161 | report_user_fault(regs, int_code, SIGSEGV, address); |
1da177e4 LT |
162 | si.si_signo = SIGSEGV; |
163 | si.si_code = si_code; | |
d2c993d8 | 164 | si.si_addr = (void __user *) address; |
1da177e4 LT |
165 | force_sig_info(SIGSEGV, &si, current); |
166 | } | |
167 | ||
50d7280d MS |
168 | static noinline void do_no_context(struct pt_regs *regs, long int_code, |
169 | unsigned long trans_exc_code) | |
10c1031f MS |
170 | { |
171 | const struct exception_table_entry *fixup; | |
61365e13 | 172 | unsigned long address; |
10c1031f MS |
173 | |
174 | /* Are we prepared to handle this kernel fault? */ | |
50d7280d | 175 | fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN); |
10c1031f MS |
176 | if (fixup) { |
177 | regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE; | |
178 | return; | |
179 | } | |
180 | ||
181 | /* | |
182 | * Oops. The kernel tried to access some bad page. We'll have to | |
183 | * terminate things with extreme prejudice. | |
184 | */ | |
61365e13 | 185 | address = trans_exc_code & __FAIL_ADDR_MASK; |
b11b5334 | 186 | if (!user_space_fault(trans_exc_code)) |
10c1031f MS |
187 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" |
188 | " at virtual kernel address %p\n", (void *)address); | |
189 | else | |
190 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
191 | " at virtual user address %p\n", (void *)address); | |
192 | ||
50d7280d | 193 | die("Oops", regs, int_code); |
10c1031f MS |
194 | do_exit(SIGKILL); |
195 | } | |
196 | ||
50d7280d MS |
197 | static noinline void do_low_address(struct pt_regs *regs, long int_code, |
198 | unsigned long trans_exc_code) | |
10c1031f MS |
199 | { |
200 | /* Low-address protection hit in kernel mode means | |
201 | NULL pointer write access in kernel mode. */ | |
202 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
203 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
50d7280d | 204 | die ("Low-address protection", regs, int_code); |
10c1031f MS |
205 | do_exit(SIGKILL); |
206 | } | |
207 | ||
50d7280d | 208 | do_no_context(regs, int_code, trans_exc_code); |
10c1031f MS |
209 | } |
210 | ||
50d7280d MS |
211 | static noinline void do_sigbus(struct pt_regs *regs, long int_code, |
212 | unsigned long trans_exc_code) | |
10c1031f MS |
213 | { |
214 | struct task_struct *tsk = current; | |
10c1031f | 215 | |
10c1031f MS |
216 | /* |
217 | * Send a sigbus, regardless of whether we were in kernel | |
218 | * or user mode. | |
219 | */ | |
61365e13 | 220 | tsk->thread.prot_addr = trans_exc_code & __FAIL_ADDR_MASK; |
50d7280d | 221 | tsk->thread.trap_no = int_code; |
10c1031f | 222 | force_sig(SIGBUS, tsk); |
10c1031f MS |
223 | } |
224 | ||
c1821c2e | 225 | #ifdef CONFIG_S390_EXEC_PROTECT |
50d7280d MS |
226 | static noinline int signal_return(struct pt_regs *regs, long int_code, |
227 | unsigned long trans_exc_code) | |
c1821c2e | 228 | { |
be5ec363 | 229 | u16 instruction; |
490f03d6 | 230 | int rc; |
be5ec363 | 231 | |
be5ec363 | 232 | rc = __get_user(instruction, (u16 __user *) regs->psw.addr); |
be5ec363 | 233 | |
50d7280d MS |
234 | if (!rc && instruction == 0x0a77) { |
235 | clear_tsk_thread_flag(current, TIF_SINGLE_STEP); | |
236 | if (is_compat_task()) | |
237 | sys32_sigreturn(); | |
238 | else | |
239 | sys_sigreturn(); | |
240 | } else if (!rc && instruction == 0x0aad) { | |
241 | clear_tsk_thread_flag(current, TIF_SINGLE_STEP); | |
242 | if (is_compat_task()) | |
243 | sys32_rt_sigreturn(); | |
244 | else | |
245 | sys_rt_sigreturn(); | |
246 | } else | |
247 | do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code); | |
c1821c2e | 248 | return 0; |
c1821c2e GS |
249 | } |
250 | #endif /* CONFIG_S390_EXEC_PROTECT */ | |
251 | ||
50d7280d MS |
252 | static noinline void do_fault_error(struct pt_regs *regs, long int_code, |
253 | unsigned long trans_exc_code, int fault) | |
254 | { | |
255 | int si_code; | |
256 | ||
257 | switch (fault) { | |
258 | case VM_FAULT_BADACCESS: | |
259 | #ifdef CONFIG_S390_EXEC_PROTECT | |
260 | if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY && | |
261 | (trans_exc_code & 3) == 0) { | |
262 | signal_return(regs, int_code, trans_exc_code); | |
263 | break; | |
264 | } | |
265 | #endif /* CONFIG_S390_EXEC_PROTECT */ | |
266 | case VM_FAULT_BADMAP: | |
267 | /* Bad memory access. Check if it is kernel or user space. */ | |
268 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
269 | /* User mode accesses just cause a SIGSEGV */ | |
270 | si_code = (fault == VM_FAULT_BADMAP) ? | |
271 | SEGV_MAPERR : SEGV_ACCERR; | |
272 | do_sigsegv(regs, int_code, si_code, trans_exc_code); | |
273 | return; | |
274 | } | |
275 | case VM_FAULT_BADCONTEXT: | |
276 | do_no_context(regs, int_code, trans_exc_code); | |
277 | break; | |
278 | default: /* fault & VM_FAULT_ERROR */ | |
279 | if (fault & VM_FAULT_OOM) | |
280 | pagefault_out_of_memory(); | |
281 | else if (fault & VM_FAULT_SIGBUS) { | |
282 | do_sigbus(regs, int_code, trans_exc_code); | |
283 | /* Kernel mode? Handle exceptions or die */ | |
284 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
285 | do_no_context(regs, int_code, trans_exc_code); | |
286 | } else | |
287 | BUG(); | |
288 | break; | |
289 | } | |
290 | } | |
291 | ||
1da177e4 LT |
292 | /* |
293 | * This routine handles page faults. It determines the address, | |
294 | * and the problem, and then passes it off to one of the appropriate | |
295 | * routines. | |
296 | * | |
50d7280d | 297 | * interruption code (int_code): |
1da177e4 LT |
298 | * 04 Protection -> Write-Protection (suprression) |
299 | * 10 Segment translation -> Not present (nullification) | |
300 | * 11 Page translation -> Not present (nullification) | |
301 | * 3b Region third trans. -> Not present (nullification) | |
302 | */ | |
1ab947de | 303 | static inline int do_exception(struct pt_regs *regs, int access, |
50d7280d | 304 | unsigned long trans_exc_code) |
1da177e4 | 305 | { |
10c1031f MS |
306 | struct task_struct *tsk; |
307 | struct mm_struct *mm; | |
308 | struct vm_area_struct *vma; | |
309 | unsigned long address; | |
92f842ea | 310 | int fault, write; |
1da177e4 | 311 | |
7ecb344a | 312 | if (notify_page_fault(regs)) |
50d7280d | 313 | return 0; |
4ba069b8 | 314 | |
10c1031f MS |
315 | tsk = current; |
316 | mm = tsk->mm; | |
1da177e4 | 317 | |
1da177e4 LT |
318 | /* |
319 | * Verify that the fault happened in user space, that | |
320 | * we are not in an interrupt and that there is a | |
321 | * user context. | |
322 | */ | |
50d7280d | 323 | fault = VM_FAULT_BADCONTEXT; |
61365e13 | 324 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
50d7280d | 325 | goto out; |
1da177e4 | 326 | |
61365e13 | 327 | address = trans_exc_code & __FAIL_ADDR_MASK; |
1da177e4 LT |
328 | /* |
329 | * When we get here, the fault happened in the current | |
330 | * task's user address space, so we can switch on the | |
331 | * interrupts again and then search the VMAs | |
332 | */ | |
333 | local_irq_enable(); | |
cdd6c482 | 334 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); |
10c1031f | 335 | down_read(&mm->mmap_sem); |
1da177e4 | 336 | |
50d7280d | 337 | fault = VM_FAULT_BADMAP; |
482b05dd GS |
338 | vma = find_vma(mm, address); |
339 | if (!vma) | |
50d7280d | 340 | goto out_up; |
c1821c2e | 341 | |
50d7280d MS |
342 | if (unlikely(vma->vm_start > address)) { |
343 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
344 | goto out_up; | |
345 | if (expand_stack(vma, address)) | |
346 | goto out_up; | |
347 | } | |
348 | ||
349 | /* | |
350 | * Ok, we have a good vm_area for this memory access, so | |
351 | * we can handle it.. | |
352 | */ | |
353 | fault = VM_FAULT_BADACCESS; | |
1ab947de | 354 | if (unlikely(!(vma->vm_flags & access))) |
50d7280d | 355 | goto out_up; |
1da177e4 | 356 | |
53492b1d GS |
357 | if (is_vm_hugetlb_page(vma)) |
358 | address &= HPAGE_MASK; | |
1da177e4 LT |
359 | /* |
360 | * If for any reason at all we couldn't handle the fault, | |
361 | * make sure we exit gracefully rather than endlessly redo | |
362 | * the fault. | |
363 | */ | |
92f842ea MS |
364 | write = (access == VM_WRITE || |
365 | (trans_exc_code & store_indication) == 0x400) ? | |
366 | FAULT_FLAG_WRITE : 0; | |
367 | fault = handle_mm_fault(mm, vma, address, write); | |
50d7280d MS |
368 | if (unlikely(fault & VM_FAULT_ERROR)) |
369 | goto out_up; | |
370 | ||
bde69af2 | 371 | if (fault & VM_FAULT_MAJOR) { |
83c54070 | 372 | tsk->maj_flt++; |
cdd6c482 | 373 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, |
bde69af2 HC |
374 | regs, address); |
375 | } else { | |
83c54070 | 376 | tsk->min_flt++; |
cdd6c482 | 377 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, |
bde69af2 HC |
378 | regs, address); |
379 | } | |
1da177e4 LT |
380 | /* |
381 | * The instruction that caused the program check will | |
382 | * be repeated. Don't signal single step via SIGTRAP. | |
383 | */ | |
482b05dd | 384 | clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP); |
50d7280d MS |
385 | fault = 0; |
386 | out_up: | |
10c1031f | 387 | up_read(&mm->mmap_sem); |
50d7280d MS |
388 | out: |
389 | return fault; | |
1da177e4 LT |
390 | } |
391 | ||
50d7280d | 392 | void __kprobes do_protection_exception(struct pt_regs *regs, long int_code) |
1da177e4 | 393 | { |
61365e13 | 394 | unsigned long trans_exc_code = S390_lowcore.trans_exc_code; |
50d7280d | 395 | int fault; |
61365e13 | 396 | |
10c1031f | 397 | /* Protection exception is supressing, decrement psw address. */ |
50d7280d | 398 | regs->psw.addr -= (int_code >> 16); |
10c1031f MS |
399 | /* |
400 | * Check for low-address protection. This needs to be treated | |
401 | * as a special case because the translation exception code | |
402 | * field is not guaranteed to contain valid data in this case. | |
403 | */ | |
61365e13 | 404 | if (unlikely(!(trans_exc_code & 4))) { |
50d7280d | 405 | do_low_address(regs, int_code, trans_exc_code); |
10c1031f MS |
406 | return; |
407 | } | |
1ab947de | 408 | fault = do_exception(regs, VM_WRITE, trans_exc_code); |
50d7280d MS |
409 | if (unlikely(fault)) |
410 | do_fault_error(regs, 4, trans_exc_code, fault); | |
1da177e4 LT |
411 | } |
412 | ||
50d7280d | 413 | void __kprobes do_dat_exception(struct pt_regs *regs, long int_code) |
1da177e4 | 414 | { |
50d7280d | 415 | unsigned long trans_exc_code = S390_lowcore.trans_exc_code; |
1ab947de | 416 | int access, fault; |
50d7280d | 417 | |
1ab947de MS |
418 | access = VM_READ | VM_EXEC | VM_WRITE; |
419 | #ifdef CONFIG_S390_EXEC_PROTECT | |
420 | if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY && | |
421 | (trans_exc_code & 3) == 0) | |
422 | access = VM_EXEC; | |
423 | #endif | |
424 | fault = do_exception(regs, access, trans_exc_code); | |
50d7280d MS |
425 | if (unlikely(fault)) |
426 | do_fault_error(regs, int_code & 255, trans_exc_code, fault); | |
1da177e4 LT |
427 | } |
428 | ||
6252d702 | 429 | #ifdef CONFIG_64BIT |
50d7280d | 430 | void __kprobes do_asce_exception(struct pt_regs *regs, long int_code) |
6252d702 | 431 | { |
61365e13 | 432 | unsigned long trans_exc_code = S390_lowcore.trans_exc_code; |
50d7280d | 433 | struct mm_struct *mm = current->mm; |
6252d702 | 434 | struct vm_area_struct *vma; |
6252d702 | 435 | |
61365e13 | 436 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
6252d702 MS |
437 | goto no_context; |
438 | ||
439 | local_irq_enable(); | |
440 | ||
441 | down_read(&mm->mmap_sem); | |
50d7280d | 442 | vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK); |
6252d702 MS |
443 | up_read(&mm->mmap_sem); |
444 | ||
445 | if (vma) { | |
446 | update_mm(mm, current); | |
447 | return; | |
448 | } | |
449 | ||
450 | /* User mode accesses just cause a SIGSEGV */ | |
451 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
50d7280d | 452 | do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code); |
6252d702 MS |
453 | return; |
454 | } | |
455 | ||
456 | no_context: | |
50d7280d | 457 | do_no_context(regs, int_code, trans_exc_code); |
6252d702 MS |
458 | } |
459 | #endif | |
460 | ||
6c1e3e79 GS |
461 | int __handle_fault(unsigned long uaddr, unsigned long int_code, int write_user) |
462 | { | |
463 | struct pt_regs regs; | |
464 | int access, fault; | |
465 | ||
466 | regs.psw.mask = psw_kernel_bits; | |
467 | if (!irqs_disabled()) | |
468 | regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT; | |
469 | regs.psw.addr = (unsigned long) __builtin_return_address(0); | |
470 | regs.psw.addr |= PSW_ADDR_AMODE; | |
471 | uaddr &= PAGE_MASK; | |
472 | access = write_user ? VM_WRITE : VM_READ; | |
473 | fault = do_exception(®s, access, uaddr | 2); | |
474 | if (unlikely(fault)) { | |
475 | if (fault & VM_FAULT_OOM) { | |
476 | pagefault_out_of_memory(); | |
477 | fault = 0; | |
478 | } else if (fault & VM_FAULT_SIGBUS) | |
479 | do_sigbus(®s, int_code, uaddr); | |
480 | } | |
481 | return fault ? -EFAULT : 0; | |
482 | } | |
483 | ||
1da177e4 LT |
484 | #ifdef CONFIG_PFAULT |
485 | /* | |
486 | * 'pfault' pseudo page faults routines. | |
487 | */ | |
29b08d2b | 488 | static ext_int_info_t ext_int_pfault; |
1da177e4 LT |
489 | static int pfault_disable = 0; |
490 | ||
491 | static int __init nopfault(char *str) | |
492 | { | |
493 | pfault_disable = 1; | |
494 | return 1; | |
495 | } | |
496 | ||
497 | __setup("nopfault", nopfault); | |
498 | ||
499 | typedef struct { | |
500 | __u16 refdiagc; | |
501 | __u16 reffcode; | |
502 | __u16 refdwlen; | |
503 | __u16 refversn; | |
504 | __u64 refgaddr; | |
505 | __u64 refselmk; | |
506 | __u64 refcmpmk; | |
507 | __u64 reserved; | |
c41fbc69 | 508 | } __attribute__ ((packed, aligned(8))) pfault_refbk_t; |
1da177e4 LT |
509 | |
510 | int pfault_init(void) | |
511 | { | |
512 | pfault_refbk_t refbk = | |
513 | { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48, | |
514 | __PF_RES_FIELD }; | |
515 | int rc; | |
516 | ||
29b08d2b | 517 | if (!MACHINE_IS_VM || pfault_disable) |
1da177e4 | 518 | return -1; |
94c12cc7 MS |
519 | asm volatile( |
520 | " diag %1,%0,0x258\n" | |
521 | "0: j 2f\n" | |
522 | "1: la %0,8\n" | |
1da177e4 | 523 | "2:\n" |
94c12cc7 MS |
524 | EX_TABLE(0b,1b) |
525 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
526 | __ctl_set_bit(0, 9); |
527 | return rc; | |
528 | } | |
529 | ||
530 | void pfault_fini(void) | |
531 | { | |
532 | pfault_refbk_t refbk = | |
533 | { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL }; | |
534 | ||
29b08d2b | 535 | if (!MACHINE_IS_VM || pfault_disable) |
1da177e4 LT |
536 | return; |
537 | __ctl_clear_bit(0,9); | |
94c12cc7 MS |
538 | asm volatile( |
539 | " diag %0,0,0x258\n" | |
1da177e4 | 540 | "0:\n" |
94c12cc7 MS |
541 | EX_TABLE(0b,0b) |
542 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
543 | } |
544 | ||
50d7280d | 545 | static void pfault_interrupt(__u16 int_code) |
1da177e4 LT |
546 | { |
547 | struct task_struct *tsk; | |
548 | __u16 subcode; | |
549 | ||
550 | /* | |
551 | * Get the external interruption subcode & pfault | |
552 | * initial/completion signal bit. VM stores this | |
553 | * in the 'cpu address' field associated with the | |
554 | * external interrupt. | |
555 | */ | |
556 | subcode = S390_lowcore.cpu_addr; | |
557 | if ((subcode & 0xff00) != __SUBCODE_MASK) | |
558 | return; | |
559 | ||
560 | /* | |
561 | * Get the token (= address of the task structure of the affected task). | |
562 | */ | |
563 | tsk = *(struct task_struct **) __LC_PFAULT_INTPARM; | |
564 | ||
565 | if (subcode & 0x0080) { | |
566 | /* signal bit is set -> a page has been swapped in by VM */ | |
567 | if (xchg(&tsk->thread.pfault_wait, -1) != 0) { | |
568 | /* Initial interrupt was faster than the completion | |
569 | * interrupt. pfault_wait is valid. Set pfault_wait | |
570 | * back to zero and wake up the process. This can | |
571 | * safely be done because the task is still sleeping | |
b6d09449 | 572 | * and can't produce new pfaults. */ |
1da177e4 LT |
573 | tsk->thread.pfault_wait = 0; |
574 | wake_up_process(tsk); | |
b6d09449 | 575 | put_task_struct(tsk); |
1da177e4 LT |
576 | } |
577 | } else { | |
578 | /* signal bit not set -> a real page is missing. */ | |
b6d09449 | 579 | get_task_struct(tsk); |
1da177e4 LT |
580 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
581 | if (xchg(&tsk->thread.pfault_wait, 1) != 0) { | |
582 | /* Completion interrupt was faster than the initial | |
583 | * interrupt (swapped in a -1 for pfault_wait). Set | |
584 | * pfault_wait back to zero and exit. This can be | |
585 | * done safely because tsk is running in kernel | |
586 | * mode and can't produce new pfaults. */ | |
587 | tsk->thread.pfault_wait = 0; | |
588 | set_task_state(tsk, TASK_RUNNING); | |
b6d09449 | 589 | put_task_struct(tsk); |
1da177e4 LT |
590 | } else |
591 | set_tsk_need_resched(tsk); | |
592 | } | |
593 | } | |
1da177e4 | 594 | |
29b08d2b HC |
595 | void __init pfault_irq_init(void) |
596 | { | |
597 | if (!MACHINE_IS_VM) | |
598 | return; | |
599 | ||
600 | /* | |
601 | * Try to get pfault pseudo page faults going. | |
602 | */ | |
603 | if (register_early_external_interrupt(0x2603, pfault_interrupt, | |
604 | &ext_int_pfault) != 0) | |
605 | panic("Couldn't request external interrupt 0x2603"); | |
606 | ||
607 | if (pfault_init() == 0) | |
608 | return; | |
609 | ||
610 | /* Tough luck, no pfault. */ | |
611 | pfault_disable = 1; | |
612 | unregister_early_external_interrupt(0x2603, pfault_interrupt, | |
613 | &ext_int_pfault); | |
614 | } | |
615 | #endif |