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> |
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 | { | |
14375bc4 | 59 | if (test_facility(2) && test_facility(75)) |
92f842ea MS |
60 | store_indication = 0xc00; |
61 | } | |
62 | ||
7ecb344a | 63 | static inline int notify_page_fault(struct pt_regs *regs) |
10c1031f | 64 | { |
33464e3b CH |
65 | int ret = 0; |
66 | ||
67 | /* kprobe_running() needs smp_processor_id() */ | |
22e0a046 | 68 | if (kprobes_built_in() && !user_mode(regs)) { |
33464e3b CH |
69 | preempt_disable(); |
70 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
71 | ret = 1; | |
72 | preempt_enable(); | |
73 | } | |
33464e3b | 74 | return ret; |
4ba069b8 | 75 | } |
4ba069b8 | 76 | |
1da177e4 LT |
77 | |
78 | /* | |
79 | * Unlock any spinlocks which will prevent us from getting the | |
cefc8be8 | 80 | * message out. |
1da177e4 LT |
81 | */ |
82 | void bust_spinlocks(int yes) | |
83 | { | |
84 | if (yes) { | |
85 | oops_in_progress = 1; | |
86 | } else { | |
87 | int loglevel_save = console_loglevel; | |
88 | console_unblank(); | |
89 | oops_in_progress = 0; | |
90 | /* | |
91 | * OK, the message is on the console. Now we call printk() | |
92 | * without oops_in_progress set so that printk will give klogd | |
93 | * a poke. Hold onto your hats... | |
94 | */ | |
95 | console_loglevel = 15; | |
96 | printk(" "); | |
97 | console_loglevel = loglevel_save; | |
98 | } | |
99 | } | |
100 | ||
101 | /* | |
482b05dd | 102 | * Returns the address space associated with the fault. |
61365e13 | 103 | * Returns 0 for kernel space and 1 for user space. |
1da177e4 | 104 | */ |
61365e13 | 105 | static inline int user_space_fault(unsigned long trans_exc_code) |
1da177e4 LT |
106 | { |
107 | /* | |
61365e13 MS |
108 | * The lowest two bits of the translation exception |
109 | * identification indicate which paging table was used. | |
1da177e4 | 110 | */ |
61365e13 MS |
111 | trans_exc_code &= 3; |
112 | if (trans_exc_code == 2) | |
113 | /* Access via secondary space, set_fs setting decides */ | |
114 | return current->thread.mm_segment.ar4; | |
b11b5334 | 115 | if (user_mode == HOME_SPACE_MODE) |
61365e13 MS |
116 | /* User space if the access has been done via home space. */ |
117 | return trans_exc_code == 3; | |
118 | /* | |
119 | * If the user space is not the home space the kernel runs in home | |
120 | * space. Access via secondary space has already been covered, | |
121 | * access via primary space or access register is from user space | |
122 | * and access via home space is from the kernel. | |
123 | */ | |
124 | return trans_exc_code != 3; | |
1da177e4 LT |
125 | } |
126 | ||
aa33c8cb | 127 | static inline void report_user_fault(struct pt_regs *regs, long signr) |
ab3c68ee HC |
128 | { |
129 | if ((task_pid_nr(current) > 1) && !show_unhandled_signals) | |
130 | return; | |
131 | if (!unhandled_signal(current, signr)) | |
132 | return; | |
133 | if (!printk_ratelimit()) | |
134 | return; | |
aa33c8cb MS |
135 | printk(KERN_ALERT "User process fault: interruption code 0x%X ", |
136 | regs->int_code); | |
ab3c68ee | 137 | print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN); |
aa33c8cb MS |
138 | printk(KERN_CONT "\n"); |
139 | printk(KERN_ALERT "failing address: %lX\n", | |
140 | regs->int_parm_long & __FAIL_ADDR_MASK); | |
ab3c68ee HC |
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 | */ | |
aa33c8cb | 148 | static noinline void do_sigsegv(struct pt_regs *regs, int si_code) |
1da177e4 LT |
149 | { |
150 | struct siginfo si; | |
151 | ||
aa33c8cb | 152 | report_user_fault(regs, SIGSEGV); |
1da177e4 LT |
153 | si.si_signo = SIGSEGV; |
154 | si.si_code = si_code; | |
aa33c8cb | 155 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
1da177e4 LT |
156 | force_sig_info(SIGSEGV, &si, current); |
157 | } | |
158 | ||
aa33c8cb | 159 | static noinline void do_no_context(struct pt_regs *regs) |
10c1031f MS |
160 | { |
161 | const struct exception_table_entry *fixup; | |
61365e13 | 162 | unsigned long address; |
10c1031f MS |
163 | |
164 | /* Are we prepared to handle this kernel fault? */ | |
50d7280d | 165 | fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN); |
10c1031f MS |
166 | if (fixup) { |
167 | regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE; | |
168 | return; | |
169 | } | |
170 | ||
171 | /* | |
172 | * Oops. The kernel tried to access some bad page. We'll have to | |
173 | * terminate things with extreme prejudice. | |
174 | */ | |
aa33c8cb MS |
175 | address = regs->int_parm_long & __FAIL_ADDR_MASK; |
176 | if (!user_space_fault(regs->int_parm_long)) | |
10c1031f MS |
177 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" |
178 | " at virtual kernel address %p\n", (void *)address); | |
179 | else | |
180 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
181 | " at virtual user address %p\n", (void *)address); | |
182 | ||
aa33c8cb | 183 | die(regs, "Oops"); |
10c1031f MS |
184 | do_exit(SIGKILL); |
185 | } | |
186 | ||
aa33c8cb | 187 | static noinline void do_low_address(struct pt_regs *regs) |
10c1031f MS |
188 | { |
189 | /* Low-address protection hit in kernel mode means | |
190 | NULL pointer write access in kernel mode. */ | |
191 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
192 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
aa33c8cb | 193 | die (regs, "Low-address protection"); |
10c1031f MS |
194 | do_exit(SIGKILL); |
195 | } | |
196 | ||
aa33c8cb | 197 | do_no_context(regs); |
10c1031f MS |
198 | } |
199 | ||
aa33c8cb | 200 | static noinline void do_sigbus(struct pt_regs *regs) |
10c1031f MS |
201 | { |
202 | struct task_struct *tsk = current; | |
36bf9680 | 203 | struct siginfo si; |
10c1031f | 204 | |
10c1031f MS |
205 | /* |
206 | * Send a sigbus, regardless of whether we were in kernel | |
207 | * or user mode. | |
208 | */ | |
36bf9680 MS |
209 | si.si_signo = SIGBUS; |
210 | si.si_errno = 0; | |
211 | si.si_code = BUS_ADRERR; | |
aa33c8cb | 212 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
36bf9680 | 213 | force_sig_info(SIGBUS, &si, tsk); |
10c1031f MS |
214 | } |
215 | ||
aa33c8cb | 216 | static noinline void do_fault_error(struct pt_regs *regs, int fault) |
50d7280d MS |
217 | { |
218 | int si_code; | |
219 | ||
220 | switch (fault) { | |
221 | case VM_FAULT_BADACCESS: | |
50d7280d MS |
222 | case VM_FAULT_BADMAP: |
223 | /* Bad memory access. Check if it is kernel or user space. */ | |
224 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
225 | /* User mode accesses just cause a SIGSEGV */ | |
226 | si_code = (fault == VM_FAULT_BADMAP) ? | |
227 | SEGV_MAPERR : SEGV_ACCERR; | |
aa33c8cb | 228 | do_sigsegv(regs, si_code); |
50d7280d MS |
229 | return; |
230 | } | |
231 | case VM_FAULT_BADCONTEXT: | |
aa33c8cb | 232 | do_no_context(regs); |
50d7280d MS |
233 | break; |
234 | default: /* fault & VM_FAULT_ERROR */ | |
99583181 HC |
235 | if (fault & VM_FAULT_OOM) { |
236 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
aa33c8cb | 237 | do_no_context(regs); |
99583181 HC |
238 | else |
239 | pagefault_out_of_memory(); | |
240 | } else if (fault & VM_FAULT_SIGBUS) { | |
50d7280d MS |
241 | /* Kernel mode? Handle exceptions or die */ |
242 | if (!(regs->psw.mask & PSW_MASK_PSTATE)) | |
aa33c8cb | 243 | do_no_context(regs); |
36bf9680 | 244 | else |
aa33c8cb | 245 | do_sigbus(regs); |
50d7280d MS |
246 | } else |
247 | BUG(); | |
248 | break; | |
249 | } | |
250 | } | |
251 | ||
1da177e4 LT |
252 | /* |
253 | * This routine handles page faults. It determines the address, | |
254 | * and the problem, and then passes it off to one of the appropriate | |
255 | * routines. | |
256 | * | |
50d7280d | 257 | * interruption code (int_code): |
1da177e4 LT |
258 | * 04 Protection -> Write-Protection (suprression) |
259 | * 10 Segment translation -> Not present (nullification) | |
260 | * 11 Page translation -> Not present (nullification) | |
261 | * 3b Region third trans. -> Not present (nullification) | |
262 | */ | |
aa33c8cb | 263 | static inline int do_exception(struct pt_regs *regs, int access) |
1da177e4 | 264 | { |
10c1031f MS |
265 | struct task_struct *tsk; |
266 | struct mm_struct *mm; | |
267 | struct vm_area_struct *vma; | |
aa33c8cb | 268 | unsigned long trans_exc_code; |
10c1031f | 269 | unsigned long address; |
33ce6140 HC |
270 | unsigned int flags; |
271 | int fault; | |
1da177e4 | 272 | |
7ecb344a | 273 | if (notify_page_fault(regs)) |
50d7280d | 274 | return 0; |
4ba069b8 | 275 | |
10c1031f MS |
276 | tsk = current; |
277 | mm = tsk->mm; | |
aa33c8cb | 278 | trans_exc_code = regs->int_parm_long; |
1da177e4 | 279 | |
1da177e4 LT |
280 | /* |
281 | * Verify that the fault happened in user space, that | |
282 | * we are not in an interrupt and that there is a | |
283 | * user context. | |
284 | */ | |
50d7280d | 285 | fault = VM_FAULT_BADCONTEXT; |
61365e13 | 286 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
50d7280d | 287 | goto out; |
1da177e4 | 288 | |
61365e13 | 289 | address = trans_exc_code & __FAIL_ADDR_MASK; |
a8b0ca17 | 290 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
33ce6140 HC |
291 | flags = FAULT_FLAG_ALLOW_RETRY; |
292 | if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400) | |
293 | flags |= FAULT_FLAG_WRITE; | |
10c1031f | 294 | down_read(&mm->mmap_sem); |
1da177e4 | 295 | |
e5992f2e MS |
296 | #ifdef CONFIG_PGSTE |
297 | if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) { | |
499069e1 | 298 | address = __gmap_fault(address, |
e5992f2e MS |
299 | (struct gmap *) S390_lowcore.gmap); |
300 | if (address == -EFAULT) { | |
301 | fault = VM_FAULT_BADMAP; | |
302 | goto out_up; | |
303 | } | |
304 | if (address == -ENOMEM) { | |
305 | fault = VM_FAULT_OOM; | |
306 | goto out_up; | |
307 | } | |
308 | } | |
309 | #endif | |
310 | ||
311 | retry: | |
50d7280d | 312 | fault = VM_FAULT_BADMAP; |
482b05dd GS |
313 | vma = find_vma(mm, address); |
314 | if (!vma) | |
50d7280d | 315 | goto out_up; |
c1821c2e | 316 | |
50d7280d MS |
317 | if (unlikely(vma->vm_start > address)) { |
318 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
319 | goto out_up; | |
320 | if (expand_stack(vma, address)) | |
321 | goto out_up; | |
322 | } | |
323 | ||
324 | /* | |
325 | * Ok, we have a good vm_area for this memory access, so | |
326 | * we can handle it.. | |
327 | */ | |
328 | fault = VM_FAULT_BADACCESS; | |
1ab947de | 329 | if (unlikely(!(vma->vm_flags & access))) |
50d7280d | 330 | goto out_up; |
1da177e4 | 331 | |
53492b1d GS |
332 | if (is_vm_hugetlb_page(vma)) |
333 | address &= HPAGE_MASK; | |
1da177e4 LT |
334 | /* |
335 | * If for any reason at all we couldn't handle the fault, | |
336 | * make sure we exit gracefully rather than endlessly redo | |
337 | * the fault. | |
338 | */ | |
33ce6140 | 339 | fault = handle_mm_fault(mm, vma, address, flags); |
50d7280d MS |
340 | if (unlikely(fault & VM_FAULT_ERROR)) |
341 | goto out_up; | |
342 | ||
33ce6140 HC |
343 | /* |
344 | * Major/minor page fault accounting is only done on the | |
345 | * initial attempt. If we go through a retry, it is extremely | |
346 | * likely that the page will be found in page cache at that point. | |
347 | */ | |
348 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
349 | if (fault & VM_FAULT_MAJOR) { | |
350 | tsk->maj_flt++; | |
a8b0ca17 | 351 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
33ce6140 HC |
352 | regs, address); |
353 | } else { | |
354 | tsk->min_flt++; | |
a8b0ca17 | 355 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
33ce6140 HC |
356 | regs, address); |
357 | } | |
358 | if (fault & VM_FAULT_RETRY) { | |
359 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
360 | * of starvation. */ | |
361 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
e5992f2e | 362 | down_read(&mm->mmap_sem); |
33ce6140 HC |
363 | goto retry; |
364 | } | |
bde69af2 | 365 | } |
1da177e4 LT |
366 | /* |
367 | * The instruction that caused the program check will | |
368 | * be repeated. Don't signal single step via SIGTRAP. | |
369 | */ | |
5e9a2692 | 370 | clear_tsk_thread_flag(tsk, TIF_PER_TRAP); |
50d7280d MS |
371 | fault = 0; |
372 | out_up: | |
10c1031f | 373 | up_read(&mm->mmap_sem); |
50d7280d MS |
374 | out: |
375 | return fault; | |
1da177e4 LT |
376 | } |
377 | ||
aa33c8cb | 378 | void __kprobes do_protection_exception(struct pt_regs *regs) |
1da177e4 | 379 | { |
aa33c8cb | 380 | unsigned long trans_exc_code; |
50d7280d | 381 | int fault; |
61365e13 | 382 | |
aa33c8cb | 383 | trans_exc_code = regs->int_parm_long; |
25985edc | 384 | /* Protection exception is suppressing, decrement psw address. */ |
aa33c8cb | 385 | regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16); |
10c1031f MS |
386 | /* |
387 | * Check for low-address protection. This needs to be treated | |
388 | * as a special case because the translation exception code | |
389 | * field is not guaranteed to contain valid data in this case. | |
390 | */ | |
61365e13 | 391 | if (unlikely(!(trans_exc_code & 4))) { |
aa33c8cb | 392 | do_low_address(regs); |
10c1031f MS |
393 | return; |
394 | } | |
aa33c8cb | 395 | fault = do_exception(regs, VM_WRITE); |
50d7280d | 396 | if (unlikely(fault)) |
aa33c8cb | 397 | do_fault_error(regs, fault); |
1da177e4 LT |
398 | } |
399 | ||
aa33c8cb | 400 | void __kprobes do_dat_exception(struct pt_regs *regs) |
1da177e4 | 401 | { |
1ab947de | 402 | int access, fault; |
50d7280d | 403 | |
1ab947de | 404 | access = VM_READ | VM_EXEC | VM_WRITE; |
aa33c8cb | 405 | fault = do_exception(regs, access); |
50d7280d | 406 | if (unlikely(fault)) |
aa33c8cb | 407 | do_fault_error(regs, fault); |
1da177e4 LT |
408 | } |
409 | ||
6252d702 | 410 | #ifdef CONFIG_64BIT |
aa33c8cb | 411 | void __kprobes do_asce_exception(struct pt_regs *regs) |
6252d702 | 412 | { |
50d7280d | 413 | struct mm_struct *mm = current->mm; |
6252d702 | 414 | struct vm_area_struct *vma; |
aa33c8cb | 415 | unsigned long trans_exc_code; |
6252d702 | 416 | |
aa33c8cb | 417 | trans_exc_code = regs->int_parm_long; |
61365e13 | 418 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
6252d702 MS |
419 | goto no_context; |
420 | ||
6252d702 | 421 | down_read(&mm->mmap_sem); |
50d7280d | 422 | vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK); |
6252d702 MS |
423 | up_read(&mm->mmap_sem); |
424 | ||
425 | if (vma) { | |
426 | update_mm(mm, current); | |
427 | return; | |
428 | } | |
429 | ||
430 | /* User mode accesses just cause a SIGSEGV */ | |
431 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
aa33c8cb | 432 | do_sigsegv(regs, SEGV_MAPERR); |
6252d702 MS |
433 | return; |
434 | } | |
435 | ||
436 | no_context: | |
aa33c8cb | 437 | do_no_context(regs); |
6252d702 MS |
438 | } |
439 | #endif | |
440 | ||
1e54622e | 441 | int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write) |
6c1e3e79 GS |
442 | { |
443 | struct pt_regs regs; | |
444 | int access, fault; | |
445 | ||
b50511e4 | 446 | regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK; |
6c1e3e79 GS |
447 | if (!irqs_disabled()) |
448 | regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT; | |
449 | regs.psw.addr = (unsigned long) __builtin_return_address(0); | |
450 | regs.psw.addr |= PSW_ADDR_AMODE; | |
aa33c8cb MS |
451 | regs.int_code = pgm_int_code; |
452 | regs.int_parm_long = (uaddr & PAGE_MASK) | 2; | |
1e54622e | 453 | access = write ? VM_WRITE : VM_READ; |
aa33c8cb | 454 | fault = do_exception(®s, access); |
6c1e3e79 | 455 | if (unlikely(fault)) { |
99583181 HC |
456 | if (fault & VM_FAULT_OOM) |
457 | return -EFAULT; | |
458 | else if (fault & VM_FAULT_SIGBUS) | |
aa33c8cb | 459 | do_sigbus(®s); |
6c1e3e79 GS |
460 | } |
461 | return fault ? -EFAULT : 0; | |
462 | } | |
463 | ||
1da177e4 LT |
464 | #ifdef CONFIG_PFAULT |
465 | /* | |
466 | * 'pfault' pseudo page faults routines. | |
467 | */ | |
fb0a9d7e | 468 | static int pfault_disable; |
1da177e4 LT |
469 | |
470 | static int __init nopfault(char *str) | |
471 | { | |
472 | pfault_disable = 1; | |
473 | return 1; | |
474 | } | |
475 | ||
476 | __setup("nopfault", nopfault); | |
477 | ||
7dd8fe1f HC |
478 | struct pfault_refbk { |
479 | u16 refdiagc; | |
480 | u16 reffcode; | |
481 | u16 refdwlen; | |
482 | u16 refversn; | |
483 | u64 refgaddr; | |
484 | u64 refselmk; | |
485 | u64 refcmpmk; | |
486 | u64 reserved; | |
487 | } __attribute__ ((packed, aligned(8))); | |
1da177e4 LT |
488 | |
489 | int pfault_init(void) | |
490 | { | |
7dd8fe1f HC |
491 | struct pfault_refbk refbk = { |
492 | .refdiagc = 0x258, | |
493 | .reffcode = 0, | |
494 | .refdwlen = 5, | |
495 | .refversn = 2, | |
496 | .refgaddr = __LC_CURRENT_PID, | |
497 | .refselmk = 1ULL << 48, | |
498 | .refcmpmk = 1ULL << 48, | |
499 | .reserved = __PF_RES_FIELD }; | |
1da177e4 LT |
500 | int rc; |
501 | ||
f32269a0 | 502 | if (pfault_disable) |
1da177e4 | 503 | return -1; |
94c12cc7 MS |
504 | asm volatile( |
505 | " diag %1,%0,0x258\n" | |
506 | "0: j 2f\n" | |
507 | "1: la %0,8\n" | |
1da177e4 | 508 | "2:\n" |
94c12cc7 MS |
509 | EX_TABLE(0b,1b) |
510 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
511 | return rc; |
512 | } | |
513 | ||
514 | void pfault_fini(void) | |
515 | { | |
7dd8fe1f HC |
516 | struct pfault_refbk refbk = { |
517 | .refdiagc = 0x258, | |
518 | .reffcode = 1, | |
519 | .refdwlen = 5, | |
520 | .refversn = 2, | |
521 | }; | |
1da177e4 | 522 | |
f32269a0 | 523 | if (pfault_disable) |
1da177e4 | 524 | return; |
94c12cc7 MS |
525 | asm volatile( |
526 | " diag %0,0,0x258\n" | |
1da177e4 | 527 | "0:\n" |
94c12cc7 MS |
528 | EX_TABLE(0b,0b) |
529 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
530 | } |
531 | ||
f2db2e6c HC |
532 | static DEFINE_SPINLOCK(pfault_lock); |
533 | static LIST_HEAD(pfault_list); | |
534 | ||
f6649a7e MS |
535 | static void pfault_interrupt(unsigned int ext_int_code, |
536 | unsigned int param32, unsigned long param64) | |
1da177e4 LT |
537 | { |
538 | struct task_struct *tsk; | |
539 | __u16 subcode; | |
f2db2e6c | 540 | pid_t pid; |
1da177e4 LT |
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 | */ | |
f6649a7e | 548 | subcode = ext_int_code >> 16; |
1da177e4 LT |
549 | if ((subcode & 0xff00) != __SUBCODE_MASK) |
550 | return; | |
a9851832 | 551 | kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++; |
f2db2e6c HC |
552 | if (subcode & 0x0080) { |
553 | /* Get the token (= pid of the affected task). */ | |
554 | pid = sizeof(void *) == 4 ? param32 : param64; | |
555 | rcu_read_lock(); | |
556 | tsk = find_task_by_pid_ns(pid, &init_pid_ns); | |
557 | if (tsk) | |
558 | get_task_struct(tsk); | |
559 | rcu_read_unlock(); | |
560 | if (!tsk) | |
561 | return; | |
562 | } else { | |
563 | tsk = current; | |
564 | } | |
565 | spin_lock(&pfault_lock); | |
1da177e4 LT |
566 | if (subcode & 0x0080) { |
567 | /* signal bit is set -> a page has been swapped in by VM */ | |
f2db2e6c | 568 | if (tsk->thread.pfault_wait == 1) { |
1da177e4 LT |
569 | /* Initial interrupt was faster than the completion |
570 | * interrupt. pfault_wait is valid. Set pfault_wait | |
571 | * back to zero and wake up the process. This can | |
572 | * safely be done because the task is still sleeping | |
b6d09449 | 573 | * and can't produce new pfaults. */ |
1da177e4 | 574 | tsk->thread.pfault_wait = 0; |
f2db2e6c | 575 | list_del(&tsk->thread.list); |
1da177e4 | 576 | wake_up_process(tsk); |
f2db2e6c HC |
577 | } else { |
578 | /* Completion interrupt was faster than initial | |
579 | * interrupt. Set pfault_wait to -1 so the initial | |
fa2fb2f4 HC |
580 | * interrupt doesn't put the task to sleep. |
581 | * If the task is not running, ignore the completion | |
582 | * interrupt since it must be a leftover of a PFAULT | |
583 | * CANCEL operation which didn't remove all pending | |
584 | * completion interrupts. */ | |
585 | if (tsk->state == TASK_RUNNING) | |
586 | tsk->thread.pfault_wait = -1; | |
1da177e4 | 587 | } |
f2db2e6c | 588 | put_task_struct(tsk); |
1da177e4 LT |
589 | } else { |
590 | /* signal bit not set -> a real page is missing. */ | |
f2db2e6c | 591 | if (tsk->thread.pfault_wait == -1) { |
1da177e4 | 592 | /* Completion interrupt was faster than the initial |
f2db2e6c HC |
593 | * interrupt (pfault_wait == -1). Set pfault_wait |
594 | * back to zero and exit. */ | |
1da177e4 | 595 | tsk->thread.pfault_wait = 0; |
f2db2e6c HC |
596 | } else { |
597 | /* Initial interrupt arrived before completion | |
598 | * interrupt. Let the task sleep. */ | |
599 | tsk->thread.pfault_wait = 1; | |
600 | list_add(&tsk->thread.list, &pfault_list); | |
601 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 602 | set_tsk_need_resched(tsk); |
f2db2e6c HC |
603 | } |
604 | } | |
605 | spin_unlock(&pfault_lock); | |
606 | } | |
607 | ||
608 | static int __cpuinit pfault_cpu_notify(struct notifier_block *self, | |
609 | unsigned long action, void *hcpu) | |
610 | { | |
611 | struct thread_struct *thread, *next; | |
612 | struct task_struct *tsk; | |
613 | ||
614 | switch (action) { | |
615 | case CPU_DEAD: | |
616 | case CPU_DEAD_FROZEN: | |
617 | spin_lock_irq(&pfault_lock); | |
618 | list_for_each_entry_safe(thread, next, &pfault_list, list) { | |
619 | thread->pfault_wait = 0; | |
620 | list_del(&thread->list); | |
621 | tsk = container_of(thread, struct task_struct, thread); | |
622 | wake_up_process(tsk); | |
623 | } | |
624 | spin_unlock_irq(&pfault_lock); | |
625 | break; | |
626 | default: | |
627 | break; | |
1da177e4 | 628 | } |
f2db2e6c | 629 | return NOTIFY_OK; |
1da177e4 | 630 | } |
1da177e4 | 631 | |
fb0a9d7e | 632 | static int __init pfault_irq_init(void) |
29b08d2b | 633 | { |
fb0a9d7e | 634 | int rc; |
29b08d2b | 635 | |
fb0a9d7e | 636 | rc = register_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
637 | if (rc) |
638 | goto out_extint; | |
639 | rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; | |
640 | if (rc) | |
641 | goto out_pfault; | |
df7997ab | 642 | service_subclass_irq_register(); |
7dd8fe1f HC |
643 | hotcpu_notifier(pfault_cpu_notify, 0); |
644 | return 0; | |
29b08d2b | 645 | |
7dd8fe1f | 646 | out_pfault: |
fb0a9d7e | 647 | unregister_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
648 | out_extint: |
649 | pfault_disable = 1; | |
650 | return rc; | |
29b08d2b | 651 | } |
fb0a9d7e HC |
652 | early_initcall(pfault_irq_init); |
653 | ||
7dd8fe1f | 654 | #endif /* CONFIG_PFAULT */ |