Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * S390 version |
a53c8fab | 3 | * Copyright IBM Corp. 1999 |
1da177e4 LT |
4 | * Author(s): Hartmut Penner (hp@de.ibm.com) |
5 | * Ulrich Weigand (uweigand@de.ibm.com) | |
6 | * | |
7 | * Derived from "arch/i386/mm/fault.c" | |
8 | * Copyright (C) 1995 Linus Torvalds | |
9 | */ | |
10 | ||
052ff461 | 11 | #include <linux/kernel_stat.h> |
cdd6c482 | 12 | #include <linux/perf_event.h> |
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> | |
7757591a | 22 | #include <linux/compat.h> |
1da177e4 | 23 | #include <linux/smp.h> |
1eeb66a1 | 24 | #include <linux/kdebug.h> |
1da177e4 LT |
25 | #include <linux/init.h> |
26 | #include <linux/console.h> | |
27 | #include <linux/module.h> | |
28 | #include <linux/hardirq.h> | |
4ba069b8 | 29 | #include <linux/kprobes.h> |
be5ec363 | 30 | #include <linux/uaccess.h> |
53492b1d | 31 | #include <linux/hugetlb.h> |
cbb870c8 | 32 | #include <asm/asm-offsets.h> |
1da177e4 | 33 | #include <asm/pgtable.h> |
d7b250e2 | 34 | #include <asm/irq.h> |
6252d702 | 35 | #include <asm/mmu_context.h> |
a0616cde | 36 | #include <asm/facility.h> |
a806170e | 37 | #include "../kernel/entry.h" |
1da177e4 | 38 | |
347a8dc3 | 39 | #ifndef CONFIG_64BIT |
1da177e4 | 40 | #define __FAIL_ADDR_MASK 0x7ffff000 |
1da177e4 LT |
41 | #define __SUBCODE_MASK 0x0200 |
42 | #define __PF_RES_FIELD 0ULL | |
347a8dc3 | 43 | #else /* CONFIG_64BIT */ |
1da177e4 | 44 | #define __FAIL_ADDR_MASK -4096L |
1da177e4 LT |
45 | #define __SUBCODE_MASK 0x0600 |
46 | #define __PF_RES_FIELD 0x8000000000000000ULL | |
347a8dc3 | 47 | #endif /* CONFIG_64BIT */ |
1da177e4 | 48 | |
50d7280d MS |
49 | #define VM_FAULT_BADCONTEXT 0x010000 |
50 | #define VM_FAULT_BADMAP 0x020000 | |
51 | #define VM_FAULT_BADACCESS 0x040000 | |
f2c76e3b | 52 | #define VM_FAULT_SIGNAL 0x080000 |
50d7280d | 53 | |
92f842ea MS |
54 | static unsigned long store_indication; |
55 | ||
56 | void fault_init(void) | |
57 | { | |
14375bc4 | 58 | if (test_facility(2) && test_facility(75)) |
92f842ea MS |
59 | store_indication = 0xc00; |
60 | } | |
61 | ||
7ecb344a | 62 | static inline int notify_page_fault(struct pt_regs *regs) |
10c1031f | 63 | { |
33464e3b CH |
64 | int ret = 0; |
65 | ||
66 | /* kprobe_running() needs smp_processor_id() */ | |
22e0a046 | 67 | if (kprobes_built_in() && !user_mode(regs)) { |
33464e3b CH |
68 | preempt_disable(); |
69 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
70 | ret = 1; | |
71 | preempt_enable(); | |
72 | } | |
33464e3b | 73 | return ret; |
4ba069b8 | 74 | } |
4ba069b8 | 75 | |
1da177e4 LT |
76 | |
77 | /* | |
78 | * Unlock any spinlocks which will prevent us from getting the | |
cefc8be8 | 79 | * message out. |
1da177e4 LT |
80 | */ |
81 | void bust_spinlocks(int yes) | |
82 | { | |
83 | if (yes) { | |
84 | oops_in_progress = 1; | |
85 | } else { | |
86 | int loglevel_save = console_loglevel; | |
87 | console_unblank(); | |
88 | oops_in_progress = 0; | |
89 | /* | |
90 | * OK, the message is on the console. Now we call printk() | |
91 | * without oops_in_progress set so that printk will give klogd | |
92 | * a poke. Hold onto your hats... | |
93 | */ | |
94 | console_loglevel = 15; | |
95 | printk(" "); | |
96 | console_loglevel = loglevel_save; | |
97 | } | |
98 | } | |
99 | ||
100 | /* | |
482b05dd | 101 | * Returns the address space associated with the fault. |
61365e13 | 102 | * Returns 0 for kernel space and 1 for user space. |
1da177e4 | 103 | */ |
61365e13 | 104 | static inline int user_space_fault(unsigned long trans_exc_code) |
1da177e4 LT |
105 | { |
106 | /* | |
61365e13 MS |
107 | * The lowest two bits of the translation exception |
108 | * identification indicate which paging table was used. | |
1da177e4 | 109 | */ |
61365e13 MS |
110 | trans_exc_code &= 3; |
111 | if (trans_exc_code == 2) | |
112 | /* Access via secondary space, set_fs setting decides */ | |
113 | return current->thread.mm_segment.ar4; | |
d1b0d842 | 114 | if (s390_user_mode == HOME_SPACE_MODE) |
61365e13 MS |
115 | /* User space if the access has been done via home space. */ |
116 | return trans_exc_code == 3; | |
117 | /* | |
118 | * If the user space is not the home space the kernel runs in home | |
119 | * space. Access via secondary space has already been covered, | |
120 | * access via primary space or access register is from user space | |
121 | * and access via home space is from the kernel. | |
122 | */ | |
123 | return trans_exc_code != 3; | |
1da177e4 LT |
124 | } |
125 | ||
aa33c8cb | 126 | static inline void report_user_fault(struct pt_regs *regs, long signr) |
ab3c68ee HC |
127 | { |
128 | if ((task_pid_nr(current) > 1) && !show_unhandled_signals) | |
129 | return; | |
130 | if (!unhandled_signal(current, signr)) | |
131 | return; | |
132 | if (!printk_ratelimit()) | |
133 | return; | |
aa33c8cb MS |
134 | printk(KERN_ALERT "User process fault: interruption code 0x%X ", |
135 | regs->int_code); | |
ab3c68ee | 136 | print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN); |
aa33c8cb MS |
137 | printk(KERN_CONT "\n"); |
138 | printk(KERN_ALERT "failing address: %lX\n", | |
139 | regs->int_parm_long & __FAIL_ADDR_MASK); | |
ab3c68ee HC |
140 | show_regs(regs); |
141 | } | |
142 | ||
1da177e4 LT |
143 | /* |
144 | * Send SIGSEGV to task. This is an external routine | |
145 | * to keep the stack usage of do_page_fault small. | |
146 | */ | |
aa33c8cb | 147 | static noinline void do_sigsegv(struct pt_regs *regs, int si_code) |
1da177e4 LT |
148 | { |
149 | struct siginfo si; | |
150 | ||
aa33c8cb | 151 | report_user_fault(regs, SIGSEGV); |
1da177e4 LT |
152 | si.si_signo = SIGSEGV; |
153 | si.si_code = si_code; | |
aa33c8cb | 154 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
1da177e4 LT |
155 | force_sig_info(SIGSEGV, &si, current); |
156 | } | |
157 | ||
aa33c8cb | 158 | static noinline void do_no_context(struct pt_regs *regs) |
10c1031f MS |
159 | { |
160 | const struct exception_table_entry *fixup; | |
61365e13 | 161 | unsigned long address; |
10c1031f MS |
162 | |
163 | /* Are we prepared to handle this kernel fault? */ | |
50d7280d | 164 | fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN); |
10c1031f | 165 | if (fixup) { |
eb608fb3 | 166 | regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE; |
10c1031f MS |
167 | return; |
168 | } | |
169 | ||
170 | /* | |
171 | * Oops. The kernel tried to access some bad page. We'll have to | |
172 | * terminate things with extreme prejudice. | |
173 | */ | |
aa33c8cb MS |
174 | address = regs->int_parm_long & __FAIL_ADDR_MASK; |
175 | if (!user_space_fault(regs->int_parm_long)) | |
10c1031f MS |
176 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" |
177 | " at virtual kernel address %p\n", (void *)address); | |
178 | else | |
179 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
180 | " at virtual user address %p\n", (void *)address); | |
181 | ||
aa33c8cb | 182 | die(regs, "Oops"); |
10c1031f MS |
183 | do_exit(SIGKILL); |
184 | } | |
185 | ||
aa33c8cb | 186 | static noinline void do_low_address(struct pt_regs *regs) |
10c1031f MS |
187 | { |
188 | /* Low-address protection hit in kernel mode means | |
189 | NULL pointer write access in kernel mode. */ | |
190 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
191 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
aa33c8cb | 192 | die (regs, "Low-address protection"); |
10c1031f MS |
193 | do_exit(SIGKILL); |
194 | } | |
195 | ||
aa33c8cb | 196 | do_no_context(regs); |
10c1031f MS |
197 | } |
198 | ||
aa33c8cb | 199 | static noinline void do_sigbus(struct pt_regs *regs) |
10c1031f MS |
200 | { |
201 | struct task_struct *tsk = current; | |
36bf9680 | 202 | struct siginfo si; |
10c1031f | 203 | |
10c1031f MS |
204 | /* |
205 | * Send a sigbus, regardless of whether we were in kernel | |
206 | * or user mode. | |
207 | */ | |
36bf9680 MS |
208 | si.si_signo = SIGBUS; |
209 | si.si_errno = 0; | |
210 | si.si_code = BUS_ADRERR; | |
aa33c8cb | 211 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
36bf9680 | 212 | force_sig_info(SIGBUS, &si, tsk); |
10c1031f MS |
213 | } |
214 | ||
aa33c8cb | 215 | static noinline void do_fault_error(struct pt_regs *regs, int fault) |
50d7280d MS |
216 | { |
217 | int si_code; | |
218 | ||
219 | switch (fault) { | |
220 | case VM_FAULT_BADACCESS: | |
50d7280d MS |
221 | case VM_FAULT_BADMAP: |
222 | /* Bad memory access. Check if it is kernel or user space. */ | |
7d256175 | 223 | if (user_mode(regs)) { |
50d7280d MS |
224 | /* User mode accesses just cause a SIGSEGV */ |
225 | si_code = (fault == VM_FAULT_BADMAP) ? | |
226 | SEGV_MAPERR : SEGV_ACCERR; | |
aa33c8cb | 227 | do_sigsegv(regs, si_code); |
50d7280d MS |
228 | return; |
229 | } | |
230 | case VM_FAULT_BADCONTEXT: | |
aa33c8cb | 231 | do_no_context(regs); |
50d7280d | 232 | break; |
f2c76e3b HC |
233 | case VM_FAULT_SIGNAL: |
234 | if (!user_mode(regs)) | |
235 | do_no_context(regs); | |
236 | break; | |
50d7280d | 237 | default: /* fault & VM_FAULT_ERROR */ |
99583181 | 238 | if (fault & VM_FAULT_OOM) { |
7d256175 | 239 | if (!user_mode(regs)) |
aa33c8cb | 240 | do_no_context(regs); |
99583181 HC |
241 | else |
242 | pagefault_out_of_memory(); | |
243 | } else if (fault & VM_FAULT_SIGBUS) { | |
50d7280d | 244 | /* Kernel mode? Handle exceptions or die */ |
7d256175 | 245 | if (!user_mode(regs)) |
aa33c8cb | 246 | do_no_context(regs); |
36bf9680 | 247 | else |
aa33c8cb | 248 | do_sigbus(regs); |
50d7280d MS |
249 | } else |
250 | BUG(); | |
251 | break; | |
252 | } | |
253 | } | |
254 | ||
1da177e4 LT |
255 | /* |
256 | * This routine handles page faults. It determines the address, | |
257 | * and the problem, and then passes it off to one of the appropriate | |
258 | * routines. | |
259 | * | |
50d7280d | 260 | * interruption code (int_code): |
1da177e4 LT |
261 | * 04 Protection -> Write-Protection (suprression) |
262 | * 10 Segment translation -> Not present (nullification) | |
263 | * 11 Page translation -> Not present (nullification) | |
264 | * 3b Region third trans. -> Not present (nullification) | |
265 | */ | |
aa33c8cb | 266 | static inline int do_exception(struct pt_regs *regs, int access) |
1da177e4 | 267 | { |
10c1031f MS |
268 | struct task_struct *tsk; |
269 | struct mm_struct *mm; | |
270 | struct vm_area_struct *vma; | |
aa33c8cb | 271 | unsigned long trans_exc_code; |
10c1031f | 272 | unsigned long address; |
33ce6140 HC |
273 | unsigned int flags; |
274 | int fault; | |
1da177e4 | 275 | |
7ecb344a | 276 | if (notify_page_fault(regs)) |
50d7280d | 277 | return 0; |
4ba069b8 | 278 | |
10c1031f MS |
279 | tsk = current; |
280 | mm = tsk->mm; | |
aa33c8cb | 281 | trans_exc_code = regs->int_parm_long; |
1da177e4 | 282 | |
1da177e4 LT |
283 | /* |
284 | * Verify that the fault happened in user space, that | |
285 | * we are not in an interrupt and that there is a | |
286 | * user context. | |
287 | */ | |
50d7280d | 288 | fault = VM_FAULT_BADCONTEXT; |
61365e13 | 289 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
50d7280d | 290 | goto out; |
1da177e4 | 291 | |
61365e13 | 292 | address = trans_exc_code & __FAIL_ADDR_MASK; |
a8b0ca17 | 293 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
f2c76e3b | 294 | flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
33ce6140 HC |
295 | if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400) |
296 | flags |= FAULT_FLAG_WRITE; | |
10c1031f | 297 | down_read(&mm->mmap_sem); |
1da177e4 | 298 | |
e5992f2e | 299 | #ifdef CONFIG_PGSTE |
5e8010cb | 300 | if ((current->flags & PF_VCPU) && S390_lowcore.gmap) { |
499069e1 | 301 | address = __gmap_fault(address, |
e5992f2e MS |
302 | (struct gmap *) S390_lowcore.gmap); |
303 | if (address == -EFAULT) { | |
304 | fault = VM_FAULT_BADMAP; | |
305 | goto out_up; | |
306 | } | |
307 | if (address == -ENOMEM) { | |
308 | fault = VM_FAULT_OOM; | |
309 | goto out_up; | |
310 | } | |
311 | } | |
312 | #endif | |
313 | ||
314 | retry: | |
50d7280d | 315 | fault = VM_FAULT_BADMAP; |
482b05dd GS |
316 | vma = find_vma(mm, address); |
317 | if (!vma) | |
50d7280d | 318 | goto out_up; |
c1821c2e | 319 | |
50d7280d MS |
320 | if (unlikely(vma->vm_start > address)) { |
321 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
322 | goto out_up; | |
323 | if (expand_stack(vma, address)) | |
324 | goto out_up; | |
325 | } | |
326 | ||
327 | /* | |
328 | * Ok, we have a good vm_area for this memory access, so | |
329 | * we can handle it.. | |
330 | */ | |
331 | fault = VM_FAULT_BADACCESS; | |
1ab947de | 332 | if (unlikely(!(vma->vm_flags & access))) |
50d7280d | 333 | goto out_up; |
1da177e4 | 334 | |
53492b1d GS |
335 | if (is_vm_hugetlb_page(vma)) |
336 | address &= HPAGE_MASK; | |
1da177e4 LT |
337 | /* |
338 | * If for any reason at all we couldn't handle the fault, | |
339 | * make sure we exit gracefully rather than endlessly redo | |
340 | * the fault. | |
341 | */ | |
33ce6140 | 342 | fault = handle_mm_fault(mm, vma, address, flags); |
f2c76e3b HC |
343 | /* No reason to continue if interrupted by SIGKILL. */ |
344 | if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) { | |
345 | fault = VM_FAULT_SIGNAL; | |
346 | goto out; | |
347 | } | |
50d7280d MS |
348 | if (unlikely(fault & VM_FAULT_ERROR)) |
349 | goto out_up; | |
350 | ||
33ce6140 HC |
351 | /* |
352 | * Major/minor page fault accounting is only done on the | |
353 | * initial attempt. If we go through a retry, it is extremely | |
354 | * likely that the page will be found in page cache at that point. | |
355 | */ | |
356 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
357 | if (fault & VM_FAULT_MAJOR) { | |
358 | tsk->maj_flt++; | |
a8b0ca17 | 359 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
33ce6140 HC |
360 | regs, address); |
361 | } else { | |
362 | tsk->min_flt++; | |
a8b0ca17 | 363 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
33ce6140 HC |
364 | regs, address); |
365 | } | |
366 | if (fault & VM_FAULT_RETRY) { | |
367 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
368 | * of starvation. */ | |
369 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
e5992f2e | 370 | down_read(&mm->mmap_sem); |
33ce6140 HC |
371 | goto retry; |
372 | } | |
bde69af2 | 373 | } |
1da177e4 LT |
374 | /* |
375 | * The instruction that caused the program check will | |
376 | * be repeated. Don't signal single step via SIGTRAP. | |
377 | */ | |
5e9a2692 | 378 | clear_tsk_thread_flag(tsk, TIF_PER_TRAP); |
50d7280d MS |
379 | fault = 0; |
380 | out_up: | |
10c1031f | 381 | up_read(&mm->mmap_sem); |
50d7280d MS |
382 | out: |
383 | return fault; | |
1da177e4 LT |
384 | } |
385 | ||
aa33c8cb | 386 | void __kprobes do_protection_exception(struct pt_regs *regs) |
1da177e4 | 387 | { |
aa33c8cb | 388 | unsigned long trans_exc_code; |
50d7280d | 389 | int fault; |
61365e13 | 390 | |
aa33c8cb | 391 | trans_exc_code = regs->int_parm_long; |
25985edc | 392 | /* Protection exception is suppressing, decrement psw address. */ |
aa33c8cb | 393 | regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16); |
10c1031f MS |
394 | /* |
395 | * Check for low-address protection. This needs to be treated | |
396 | * as a special case because the translation exception code | |
397 | * field is not guaranteed to contain valid data in this case. | |
398 | */ | |
61365e13 | 399 | if (unlikely(!(trans_exc_code & 4))) { |
aa33c8cb | 400 | do_low_address(regs); |
10c1031f MS |
401 | return; |
402 | } | |
aa33c8cb | 403 | fault = do_exception(regs, VM_WRITE); |
50d7280d | 404 | if (unlikely(fault)) |
aa33c8cb | 405 | do_fault_error(regs, fault); |
1da177e4 LT |
406 | } |
407 | ||
aa33c8cb | 408 | void __kprobes do_dat_exception(struct pt_regs *regs) |
1da177e4 | 409 | { |
1ab947de | 410 | int access, fault; |
50d7280d | 411 | |
1ab947de | 412 | access = VM_READ | VM_EXEC | VM_WRITE; |
aa33c8cb | 413 | fault = do_exception(regs, access); |
50d7280d | 414 | if (unlikely(fault)) |
aa33c8cb | 415 | do_fault_error(regs, fault); |
1da177e4 LT |
416 | } |
417 | ||
6252d702 | 418 | #ifdef CONFIG_64BIT |
aa33c8cb | 419 | void __kprobes do_asce_exception(struct pt_regs *regs) |
6252d702 | 420 | { |
50d7280d | 421 | struct mm_struct *mm = current->mm; |
6252d702 | 422 | struct vm_area_struct *vma; |
aa33c8cb | 423 | unsigned long trans_exc_code; |
6252d702 | 424 | |
aa33c8cb | 425 | trans_exc_code = regs->int_parm_long; |
61365e13 | 426 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
6252d702 MS |
427 | goto no_context; |
428 | ||
6252d702 | 429 | down_read(&mm->mmap_sem); |
50d7280d | 430 | vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK); |
6252d702 MS |
431 | up_read(&mm->mmap_sem); |
432 | ||
433 | if (vma) { | |
434 | update_mm(mm, current); | |
435 | return; | |
436 | } | |
437 | ||
438 | /* User mode accesses just cause a SIGSEGV */ | |
7d256175 | 439 | if (user_mode(regs)) { |
aa33c8cb | 440 | do_sigsegv(regs, SEGV_MAPERR); |
6252d702 MS |
441 | return; |
442 | } | |
443 | ||
444 | no_context: | |
aa33c8cb | 445 | do_no_context(regs); |
6252d702 MS |
446 | } |
447 | #endif | |
448 | ||
1e54622e | 449 | int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write) |
6c1e3e79 GS |
450 | { |
451 | struct pt_regs regs; | |
452 | int access, fault; | |
453 | ||
008c2e8f | 454 | /* Emulate a uaccess fault from kernel mode. */ |
b50511e4 | 455 | regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK; |
6c1e3e79 GS |
456 | if (!irqs_disabled()) |
457 | regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT; | |
458 | regs.psw.addr = (unsigned long) __builtin_return_address(0); | |
459 | regs.psw.addr |= PSW_ADDR_AMODE; | |
aa33c8cb MS |
460 | regs.int_code = pgm_int_code; |
461 | regs.int_parm_long = (uaddr & PAGE_MASK) | 2; | |
1e54622e | 462 | access = write ? VM_WRITE : VM_READ; |
aa33c8cb | 463 | fault = do_exception(®s, access); |
008c2e8f HC |
464 | /* |
465 | * Since the fault happened in kernel mode while performing a uaccess | |
466 | * all we need to do now is emulating a fixup in case "fault" is not | |
467 | * zero. | |
468 | * For the calling uaccess functions this results always in -EFAULT. | |
469 | */ | |
6c1e3e79 GS |
470 | return fault ? -EFAULT : 0; |
471 | } | |
472 | ||
1da177e4 LT |
473 | #ifdef CONFIG_PFAULT |
474 | /* | |
475 | * 'pfault' pseudo page faults routines. | |
476 | */ | |
fb0a9d7e | 477 | static int pfault_disable; |
1da177e4 LT |
478 | |
479 | static int __init nopfault(char *str) | |
480 | { | |
481 | pfault_disable = 1; | |
482 | return 1; | |
483 | } | |
484 | ||
485 | __setup("nopfault", nopfault); | |
486 | ||
7dd8fe1f HC |
487 | struct pfault_refbk { |
488 | u16 refdiagc; | |
489 | u16 reffcode; | |
490 | u16 refdwlen; | |
491 | u16 refversn; | |
492 | u64 refgaddr; | |
493 | u64 refselmk; | |
494 | u64 refcmpmk; | |
495 | u64 reserved; | |
496 | } __attribute__ ((packed, aligned(8))); | |
1da177e4 LT |
497 | |
498 | int pfault_init(void) | |
499 | { | |
7dd8fe1f HC |
500 | struct pfault_refbk refbk = { |
501 | .refdiagc = 0x258, | |
502 | .reffcode = 0, | |
503 | .refdwlen = 5, | |
504 | .refversn = 2, | |
505 | .refgaddr = __LC_CURRENT_PID, | |
506 | .refselmk = 1ULL << 48, | |
507 | .refcmpmk = 1ULL << 48, | |
508 | .reserved = __PF_RES_FIELD }; | |
1da177e4 LT |
509 | int rc; |
510 | ||
f32269a0 | 511 | if (pfault_disable) |
1da177e4 | 512 | return -1; |
94c12cc7 MS |
513 | asm volatile( |
514 | " diag %1,%0,0x258\n" | |
515 | "0: j 2f\n" | |
516 | "1: la %0,8\n" | |
1da177e4 | 517 | "2:\n" |
94c12cc7 MS |
518 | EX_TABLE(0b,1b) |
519 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
520 | return rc; |
521 | } | |
522 | ||
523 | void pfault_fini(void) | |
524 | { | |
7dd8fe1f HC |
525 | struct pfault_refbk refbk = { |
526 | .refdiagc = 0x258, | |
527 | .reffcode = 1, | |
528 | .refdwlen = 5, | |
529 | .refversn = 2, | |
530 | }; | |
1da177e4 | 531 | |
f32269a0 | 532 | if (pfault_disable) |
1da177e4 | 533 | return; |
94c12cc7 MS |
534 | asm volatile( |
535 | " diag %0,0,0x258\n" | |
1da177e4 | 536 | "0:\n" |
94c12cc7 MS |
537 | EX_TABLE(0b,0b) |
538 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
539 | } |
540 | ||
f2db2e6c HC |
541 | static DEFINE_SPINLOCK(pfault_lock); |
542 | static LIST_HEAD(pfault_list); | |
543 | ||
fde15c3a | 544 | static void pfault_interrupt(struct ext_code ext_code, |
f6649a7e | 545 | unsigned int param32, unsigned long param64) |
1da177e4 LT |
546 | { |
547 | struct task_struct *tsk; | |
548 | __u16 subcode; | |
f2db2e6c | 549 | pid_t pid; |
1da177e4 LT |
550 | |
551 | /* | |
552 | * Get the external interruption subcode & pfault | |
553 | * initial/completion signal bit. VM stores this | |
554 | * in the 'cpu address' field associated with the | |
555 | * external interrupt. | |
556 | */ | |
fde15c3a | 557 | subcode = ext_code.subcode; |
1da177e4 LT |
558 | if ((subcode & 0xff00) != __SUBCODE_MASK) |
559 | return; | |
a9851832 | 560 | kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++; |
54c27791 HC |
561 | /* Get the token (= pid of the affected task). */ |
562 | pid = sizeof(void *) == 4 ? param32 : param64; | |
563 | rcu_read_lock(); | |
564 | tsk = find_task_by_pid_ns(pid, &init_pid_ns); | |
565 | if (tsk) | |
566 | get_task_struct(tsk); | |
567 | rcu_read_unlock(); | |
568 | if (!tsk) | |
569 | return; | |
f2db2e6c | 570 | spin_lock(&pfault_lock); |
1da177e4 LT |
571 | if (subcode & 0x0080) { |
572 | /* signal bit is set -> a page has been swapped in by VM */ | |
f2db2e6c | 573 | if (tsk->thread.pfault_wait == 1) { |
1da177e4 LT |
574 | /* Initial interrupt was faster than the completion |
575 | * interrupt. pfault_wait is valid. Set pfault_wait | |
576 | * back to zero and wake up the process. This can | |
577 | * safely be done because the task is still sleeping | |
b6d09449 | 578 | * and can't produce new pfaults. */ |
1da177e4 | 579 | tsk->thread.pfault_wait = 0; |
f2db2e6c | 580 | list_del(&tsk->thread.list); |
1da177e4 | 581 | wake_up_process(tsk); |
d5e50a51 | 582 | put_task_struct(tsk); |
f2db2e6c HC |
583 | } else { |
584 | /* Completion interrupt was faster than initial | |
585 | * interrupt. Set pfault_wait to -1 so the initial | |
fa2fb2f4 HC |
586 | * interrupt doesn't put the task to sleep. |
587 | * If the task is not running, ignore the completion | |
588 | * interrupt since it must be a leftover of a PFAULT | |
589 | * CANCEL operation which didn't remove all pending | |
590 | * completion interrupts. */ | |
591 | if (tsk->state == TASK_RUNNING) | |
592 | tsk->thread.pfault_wait = -1; | |
1da177e4 LT |
593 | } |
594 | } else { | |
595 | /* signal bit not set -> a real page is missing. */ | |
d49f47f8 HC |
596 | if (WARN_ON_ONCE(tsk != current)) |
597 | goto out; | |
d5e50a51 HC |
598 | if (tsk->thread.pfault_wait == 1) { |
599 | /* Already on the list with a reference: put to sleep */ | |
0a16ba78 | 600 | __set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
d5e50a51 HC |
601 | set_tsk_need_resched(tsk); |
602 | } else if (tsk->thread.pfault_wait == -1) { | |
1da177e4 | 603 | /* Completion interrupt was faster than the initial |
f2db2e6c HC |
604 | * interrupt (pfault_wait == -1). Set pfault_wait |
605 | * back to zero and exit. */ | |
1da177e4 | 606 | tsk->thread.pfault_wait = 0; |
f2db2e6c HC |
607 | } else { |
608 | /* Initial interrupt arrived before completion | |
d5e50a51 HC |
609 | * interrupt. Let the task sleep. |
610 | * An extra task reference is needed since a different | |
611 | * cpu may set the task state to TASK_RUNNING again | |
612 | * before the scheduler is reached. */ | |
613 | get_task_struct(tsk); | |
f2db2e6c HC |
614 | tsk->thread.pfault_wait = 1; |
615 | list_add(&tsk->thread.list, &pfault_list); | |
0a16ba78 | 616 | __set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
1da177e4 | 617 | set_tsk_need_resched(tsk); |
f2db2e6c HC |
618 | } |
619 | } | |
d49f47f8 | 620 | out: |
f2db2e6c | 621 | spin_unlock(&pfault_lock); |
54c27791 | 622 | put_task_struct(tsk); |
f2db2e6c HC |
623 | } |
624 | ||
625 | static int __cpuinit pfault_cpu_notify(struct notifier_block *self, | |
626 | unsigned long action, void *hcpu) | |
627 | { | |
628 | struct thread_struct *thread, *next; | |
629 | struct task_struct *tsk; | |
630 | ||
1c725922 | 631 | switch (action & ~CPU_TASKS_FROZEN) { |
f2db2e6c | 632 | case CPU_DEAD: |
f2db2e6c HC |
633 | spin_lock_irq(&pfault_lock); |
634 | list_for_each_entry_safe(thread, next, &pfault_list, list) { | |
635 | thread->pfault_wait = 0; | |
636 | list_del(&thread->list); | |
637 | tsk = container_of(thread, struct task_struct, thread); | |
638 | wake_up_process(tsk); | |
d5e50a51 | 639 | put_task_struct(tsk); |
f2db2e6c HC |
640 | } |
641 | spin_unlock_irq(&pfault_lock); | |
642 | break; | |
643 | default: | |
644 | break; | |
1da177e4 | 645 | } |
f2db2e6c | 646 | return NOTIFY_OK; |
1da177e4 | 647 | } |
1da177e4 | 648 | |
fb0a9d7e | 649 | static int __init pfault_irq_init(void) |
29b08d2b | 650 | { |
fb0a9d7e | 651 | int rc; |
29b08d2b | 652 | |
fb0a9d7e | 653 | rc = register_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
654 | if (rc) |
655 | goto out_extint; | |
656 | rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; | |
657 | if (rc) | |
658 | goto out_pfault; | |
df7997ab | 659 | service_subclass_irq_register(); |
7dd8fe1f HC |
660 | hotcpu_notifier(pfault_cpu_notify, 0); |
661 | return 0; | |
29b08d2b | 662 | |
7dd8fe1f | 663 | out_pfault: |
fb0a9d7e | 664 | unregister_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
665 | out_extint: |
666 | pfault_disable = 1; | |
667 | return rc; | |
29b08d2b | 668 | } |
fb0a9d7e HC |
669 | early_initcall(pfault_irq_init); |
670 | ||
7dd8fe1f | 671 | #endif /* CONFIG_PFAULT */ |