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 | |
a4f32bdb | 52 | #define VM_FAULT_SIGNAL 0x080000 |
50d7280d | 53 | |
a4f32bdb | 54 | static unsigned long store_indication __read_mostly; |
92f842ea | 55 | |
a4f32bdb HC |
56 | #ifdef CONFIG_64BIT |
57 | static int __init fault_init(void) | |
92f842ea | 58 | { |
a4f32bdb | 59 | if (test_facility(75)) |
92f842ea | 60 | store_indication = 0xc00; |
a4f32bdb | 61 | return 0; |
92f842ea | 62 | } |
a4f32bdb HC |
63 | early_initcall(fault_init); |
64 | #endif | |
92f842ea | 65 | |
7ecb344a | 66 | static inline int notify_page_fault(struct pt_regs *regs) |
10c1031f | 67 | { |
33464e3b CH |
68 | int ret = 0; |
69 | ||
70 | /* kprobe_running() needs smp_processor_id() */ | |
22e0a046 | 71 | if (kprobes_built_in() && !user_mode(regs)) { |
33464e3b CH |
72 | preempt_disable(); |
73 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
74 | ret = 1; | |
75 | preempt_enable(); | |
76 | } | |
33464e3b | 77 | return ret; |
4ba069b8 | 78 | } |
4ba069b8 | 79 | |
1da177e4 LT |
80 | |
81 | /* | |
82 | * Unlock any spinlocks which will prevent us from getting the | |
cefc8be8 | 83 | * message out. |
1da177e4 LT |
84 | */ |
85 | void bust_spinlocks(int yes) | |
86 | { | |
87 | if (yes) { | |
88 | oops_in_progress = 1; | |
89 | } else { | |
90 | int loglevel_save = console_loglevel; | |
91 | console_unblank(); | |
92 | oops_in_progress = 0; | |
93 | /* | |
94 | * OK, the message is on the console. Now we call printk() | |
95 | * without oops_in_progress set so that printk will give klogd | |
96 | * a poke. Hold onto your hats... | |
97 | */ | |
98 | console_loglevel = 15; | |
99 | printk(" "); | |
100 | console_loglevel = loglevel_save; | |
101 | } | |
102 | } | |
103 | ||
104 | /* | |
482b05dd | 105 | * Returns the address space associated with the fault. |
61365e13 | 106 | * Returns 0 for kernel space and 1 for user space. |
1da177e4 | 107 | */ |
61365e13 | 108 | static inline int user_space_fault(unsigned long trans_exc_code) |
1da177e4 LT |
109 | { |
110 | /* | |
61365e13 MS |
111 | * The lowest two bits of the translation exception |
112 | * identification indicate which paging table was used. | |
1da177e4 | 113 | */ |
61365e13 MS |
114 | trans_exc_code &= 3; |
115 | if (trans_exc_code == 2) | |
116 | /* Access via secondary space, set_fs setting decides */ | |
117 | return current->thread.mm_segment.ar4; | |
61365e13 | 118 | /* |
e258d719 | 119 | * Access via primary space or access register is from user space |
61365e13 MS |
120 | * and access via home space is from the kernel. |
121 | */ | |
122 | return trans_exc_code != 3; | |
1da177e4 LT |
123 | } |
124 | ||
aa33c8cb | 125 | static inline void report_user_fault(struct pt_regs *regs, long signr) |
ab3c68ee HC |
126 | { |
127 | if ((task_pid_nr(current) > 1) && !show_unhandled_signals) | |
128 | return; | |
129 | if (!unhandled_signal(current, signr)) | |
130 | return; | |
131 | if (!printk_ratelimit()) | |
132 | return; | |
aa33c8cb MS |
133 | printk(KERN_ALERT "User process fault: interruption code 0x%X ", |
134 | regs->int_code); | |
ab3c68ee | 135 | print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN); |
aa33c8cb MS |
136 | printk(KERN_CONT "\n"); |
137 | printk(KERN_ALERT "failing address: %lX\n", | |
138 | regs->int_parm_long & __FAIL_ADDR_MASK); | |
ab3c68ee HC |
139 | show_regs(regs); |
140 | } | |
141 | ||
1da177e4 LT |
142 | /* |
143 | * Send SIGSEGV to task. This is an external routine | |
144 | * to keep the stack usage of do_page_fault small. | |
145 | */ | |
aa33c8cb | 146 | static noinline void do_sigsegv(struct pt_regs *regs, int si_code) |
1da177e4 LT |
147 | { |
148 | struct siginfo si; | |
149 | ||
aa33c8cb | 150 | report_user_fault(regs, SIGSEGV); |
1da177e4 LT |
151 | si.si_signo = SIGSEGV; |
152 | si.si_code = si_code; | |
aa33c8cb | 153 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
1da177e4 LT |
154 | force_sig_info(SIGSEGV, &si, current); |
155 | } | |
156 | ||
aa33c8cb | 157 | static noinline void do_no_context(struct pt_regs *regs) |
10c1031f MS |
158 | { |
159 | const struct exception_table_entry *fixup; | |
61365e13 | 160 | unsigned long address; |
10c1031f MS |
161 | |
162 | /* Are we prepared to handle this kernel fault? */ | |
50d7280d | 163 | fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN); |
10c1031f | 164 | if (fixup) { |
eb608fb3 | 165 | regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE; |
10c1031f MS |
166 | return; |
167 | } | |
168 | ||
169 | /* | |
170 | * Oops. The kernel tried to access some bad page. We'll have to | |
171 | * terminate things with extreme prejudice. | |
172 | */ | |
aa33c8cb MS |
173 | address = regs->int_parm_long & __FAIL_ADDR_MASK; |
174 | if (!user_space_fault(regs->int_parm_long)) | |
10c1031f MS |
175 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" |
176 | " at virtual kernel address %p\n", (void *)address); | |
177 | else | |
178 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
179 | " at virtual user address %p\n", (void *)address); | |
180 | ||
aa33c8cb | 181 | die(regs, "Oops"); |
10c1031f MS |
182 | do_exit(SIGKILL); |
183 | } | |
184 | ||
aa33c8cb | 185 | static noinline void do_low_address(struct pt_regs *regs) |
10c1031f MS |
186 | { |
187 | /* Low-address protection hit in kernel mode means | |
188 | NULL pointer write access in kernel mode. */ | |
189 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
190 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
aa33c8cb | 191 | die (regs, "Low-address protection"); |
10c1031f MS |
192 | do_exit(SIGKILL); |
193 | } | |
194 | ||
aa33c8cb | 195 | do_no_context(regs); |
10c1031f MS |
196 | } |
197 | ||
aa33c8cb | 198 | static noinline void do_sigbus(struct pt_regs *regs) |
10c1031f MS |
199 | { |
200 | struct task_struct *tsk = current; | |
36bf9680 | 201 | struct siginfo si; |
10c1031f | 202 | |
10c1031f MS |
203 | /* |
204 | * Send a sigbus, regardless of whether we were in kernel | |
205 | * or user mode. | |
206 | */ | |
36bf9680 MS |
207 | si.si_signo = SIGBUS; |
208 | si.si_errno = 0; | |
209 | si.si_code = BUS_ADRERR; | |
aa33c8cb | 210 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
36bf9680 | 211 | force_sig_info(SIGBUS, &si, tsk); |
10c1031f MS |
212 | } |
213 | ||
aa33c8cb | 214 | static noinline void do_fault_error(struct pt_regs *regs, int fault) |
50d7280d MS |
215 | { |
216 | int si_code; | |
217 | ||
218 | switch (fault) { | |
219 | case VM_FAULT_BADACCESS: | |
50d7280d MS |
220 | case VM_FAULT_BADMAP: |
221 | /* Bad memory access. Check if it is kernel or user space. */ | |
7d256175 | 222 | if (user_mode(regs)) { |
50d7280d MS |
223 | /* User mode accesses just cause a SIGSEGV */ |
224 | si_code = (fault == VM_FAULT_BADMAP) ? | |
225 | SEGV_MAPERR : SEGV_ACCERR; | |
aa33c8cb | 226 | do_sigsegv(regs, si_code); |
50d7280d MS |
227 | return; |
228 | } | |
229 | case VM_FAULT_BADCONTEXT: | |
aa33c8cb | 230 | do_no_context(regs); |
50d7280d | 231 | break; |
f2c76e3b HC |
232 | case VM_FAULT_SIGNAL: |
233 | if (!user_mode(regs)) | |
234 | do_no_context(regs); | |
235 | break; | |
50d7280d | 236 | default: /* fault & VM_FAULT_ERROR */ |
99583181 | 237 | if (fault & VM_FAULT_OOM) { |
7d256175 | 238 | if (!user_mode(regs)) |
aa33c8cb | 239 | do_no_context(regs); |
99583181 HC |
240 | else |
241 | pagefault_out_of_memory(); | |
242 | } else if (fault & VM_FAULT_SIGBUS) { | |
50d7280d | 243 | /* Kernel mode? Handle exceptions or die */ |
7d256175 | 244 | if (!user_mode(regs)) |
aa33c8cb | 245 | do_no_context(regs); |
36bf9680 | 246 | else |
aa33c8cb | 247 | do_sigbus(regs); |
50d7280d MS |
248 | } else |
249 | BUG(); | |
250 | break; | |
251 | } | |
252 | } | |
253 | ||
1da177e4 LT |
254 | /* |
255 | * This routine handles page faults. It determines the address, | |
256 | * and the problem, and then passes it off to one of the appropriate | |
257 | * routines. | |
258 | * | |
50d7280d | 259 | * interruption code (int_code): |
1da177e4 LT |
260 | * 04 Protection -> Write-Protection (suprression) |
261 | * 10 Segment translation -> Not present (nullification) | |
262 | * 11 Page translation -> Not present (nullification) | |
263 | * 3b Region third trans. -> Not present (nullification) | |
264 | */ | |
aa33c8cb | 265 | static inline int do_exception(struct pt_regs *regs, int access) |
1da177e4 | 266 | { |
10c1031f MS |
267 | struct task_struct *tsk; |
268 | struct mm_struct *mm; | |
269 | struct vm_area_struct *vma; | |
aa33c8cb | 270 | unsigned long trans_exc_code; |
10c1031f | 271 | unsigned long address; |
33ce6140 HC |
272 | unsigned int flags; |
273 | int fault; | |
1da177e4 | 274 | |
39efd4ec MS |
275 | tsk = current; |
276 | /* | |
277 | * The instruction that caused the program check has | |
278 | * been nullified. Don't signal single step via SIGTRAP. | |
279 | */ | |
280 | clear_tsk_thread_flag(tsk, TIF_PER_TRAP); | |
281 | ||
7ecb344a | 282 | if (notify_page_fault(regs)) |
50d7280d | 283 | return 0; |
4ba069b8 | 284 | |
10c1031f | 285 | mm = tsk->mm; |
aa33c8cb | 286 | trans_exc_code = regs->int_parm_long; |
1da177e4 | 287 | |
1da177e4 LT |
288 | /* |
289 | * Verify that the fault happened in user space, that | |
290 | * we are not in an interrupt and that there is a | |
291 | * user context. | |
292 | */ | |
50d7280d | 293 | fault = VM_FAULT_BADCONTEXT; |
61365e13 | 294 | if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm)) |
50d7280d | 295 | goto out; |
1da177e4 | 296 | |
61365e13 | 297 | address = trans_exc_code & __FAIL_ADDR_MASK; |
a8b0ca17 | 298 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
f2c76e3b | 299 | flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
759496ba JW |
300 | if (user_mode(regs)) |
301 | flags |= FAULT_FLAG_USER; | |
33ce6140 HC |
302 | if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400) |
303 | flags |= FAULT_FLAG_WRITE; | |
10c1031f | 304 | down_read(&mm->mmap_sem); |
1da177e4 | 305 | |
e5992f2e | 306 | #ifdef CONFIG_PGSTE |
5e8010cb | 307 | if ((current->flags & PF_VCPU) && S390_lowcore.gmap) { |
499069e1 | 308 | address = __gmap_fault(address, |
e5992f2e MS |
309 | (struct gmap *) S390_lowcore.gmap); |
310 | if (address == -EFAULT) { | |
311 | fault = VM_FAULT_BADMAP; | |
312 | goto out_up; | |
313 | } | |
314 | if (address == -ENOMEM) { | |
315 | fault = VM_FAULT_OOM; | |
316 | goto out_up; | |
317 | } | |
318 | } | |
319 | #endif | |
320 | ||
321 | retry: | |
50d7280d | 322 | fault = VM_FAULT_BADMAP; |
482b05dd GS |
323 | vma = find_vma(mm, address); |
324 | if (!vma) | |
50d7280d | 325 | goto out_up; |
c1821c2e | 326 | |
50d7280d MS |
327 | if (unlikely(vma->vm_start > address)) { |
328 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
329 | goto out_up; | |
330 | if (expand_stack(vma, address)) | |
331 | goto out_up; | |
332 | } | |
333 | ||
334 | /* | |
335 | * Ok, we have a good vm_area for this memory access, so | |
336 | * we can handle it.. | |
337 | */ | |
338 | fault = VM_FAULT_BADACCESS; | |
1ab947de | 339 | if (unlikely(!(vma->vm_flags & access))) |
50d7280d | 340 | goto out_up; |
1da177e4 | 341 | |
53492b1d GS |
342 | if (is_vm_hugetlb_page(vma)) |
343 | address &= HPAGE_MASK; | |
1da177e4 LT |
344 | /* |
345 | * If for any reason at all we couldn't handle the fault, | |
346 | * make sure we exit gracefully rather than endlessly redo | |
347 | * the fault. | |
348 | */ | |
33ce6140 | 349 | fault = handle_mm_fault(mm, vma, address, flags); |
f2c76e3b HC |
350 | /* No reason to continue if interrupted by SIGKILL. */ |
351 | if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) { | |
352 | fault = VM_FAULT_SIGNAL; | |
353 | goto out; | |
354 | } | |
50d7280d MS |
355 | if (unlikely(fault & VM_FAULT_ERROR)) |
356 | goto out_up; | |
357 | ||
33ce6140 HC |
358 | /* |
359 | * Major/minor page fault accounting is only done on the | |
360 | * initial attempt. If we go through a retry, it is extremely | |
361 | * likely that the page will be found in page cache at that point. | |
362 | */ | |
363 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
364 | if (fault & VM_FAULT_MAJOR) { | |
365 | tsk->maj_flt++; | |
a8b0ca17 | 366 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
33ce6140 HC |
367 | regs, address); |
368 | } else { | |
369 | tsk->min_flt++; | |
a8b0ca17 | 370 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
33ce6140 HC |
371 | regs, address); |
372 | } | |
373 | if (fault & VM_FAULT_RETRY) { | |
374 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
375 | * of starvation. */ | |
376 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
45cac65b | 377 | flags |= FAULT_FLAG_TRIED; |
e5992f2e | 378 | down_read(&mm->mmap_sem); |
33ce6140 HC |
379 | goto retry; |
380 | } | |
bde69af2 | 381 | } |
50d7280d MS |
382 | fault = 0; |
383 | out_up: | |
10c1031f | 384 | up_read(&mm->mmap_sem); |
50d7280d MS |
385 | out: |
386 | return fault; | |
1da177e4 LT |
387 | } |
388 | ||
aa33c8cb | 389 | void __kprobes do_protection_exception(struct pt_regs *regs) |
1da177e4 | 390 | { |
aa33c8cb | 391 | unsigned long trans_exc_code; |
50d7280d | 392 | int fault; |
61365e13 | 393 | |
aa33c8cb | 394 | trans_exc_code = regs->int_parm_long; |
f752ac4d MS |
395 | /* |
396 | * Protection exceptions are suppressing, decrement psw address. | |
397 | * The exception to this rule are aborted transactions, for these | |
398 | * the PSW already points to the correct location. | |
399 | */ | |
400 | if (!(regs->int_code & 0x200)) | |
401 | regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16); | |
10c1031f MS |
402 | /* |
403 | * Check for low-address protection. This needs to be treated | |
404 | * as a special case because the translation exception code | |
405 | * field is not guaranteed to contain valid data in this case. | |
406 | */ | |
61365e13 | 407 | if (unlikely(!(trans_exc_code & 4))) { |
aa33c8cb | 408 | do_low_address(regs); |
10c1031f MS |
409 | return; |
410 | } | |
aa33c8cb | 411 | fault = do_exception(regs, VM_WRITE); |
50d7280d | 412 | if (unlikely(fault)) |
aa33c8cb | 413 | do_fault_error(regs, fault); |
1da177e4 LT |
414 | } |
415 | ||
aa33c8cb | 416 | void __kprobes do_dat_exception(struct pt_regs *regs) |
1da177e4 | 417 | { |
1ab947de | 418 | int access, fault; |
50d7280d | 419 | |
1ab947de | 420 | access = VM_READ | VM_EXEC | VM_WRITE; |
aa33c8cb | 421 | fault = do_exception(regs, access); |
50d7280d | 422 | if (unlikely(fault)) |
aa33c8cb | 423 | do_fault_error(regs, fault); |
1da177e4 LT |
424 | } |
425 | ||
1e54622e | 426 | int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write) |
6c1e3e79 GS |
427 | { |
428 | struct pt_regs regs; | |
429 | int access, fault; | |
430 | ||
008c2e8f | 431 | /* Emulate a uaccess fault from kernel mode. */ |
e258d719 | 432 | regs.psw.mask = PSW_KERNEL_BITS | PSW_MASK_DAT | PSW_MASK_MCHECK; |
6c1e3e79 GS |
433 | if (!irqs_disabled()) |
434 | regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT; | |
435 | regs.psw.addr = (unsigned long) __builtin_return_address(0); | |
436 | regs.psw.addr |= PSW_ADDR_AMODE; | |
aa33c8cb MS |
437 | regs.int_code = pgm_int_code; |
438 | regs.int_parm_long = (uaddr & PAGE_MASK) | 2; | |
1e54622e | 439 | access = write ? VM_WRITE : VM_READ; |
aa33c8cb | 440 | fault = do_exception(®s, access); |
008c2e8f HC |
441 | /* |
442 | * Since the fault happened in kernel mode while performing a uaccess | |
443 | * all we need to do now is emulating a fixup in case "fault" is not | |
444 | * zero. | |
445 | * For the calling uaccess functions this results always in -EFAULT. | |
446 | */ | |
6c1e3e79 GS |
447 | return fault ? -EFAULT : 0; |
448 | } | |
449 | ||
1da177e4 LT |
450 | #ifdef CONFIG_PFAULT |
451 | /* | |
452 | * 'pfault' pseudo page faults routines. | |
453 | */ | |
fb0a9d7e | 454 | static int pfault_disable; |
1da177e4 LT |
455 | |
456 | static int __init nopfault(char *str) | |
457 | { | |
458 | pfault_disable = 1; | |
459 | return 1; | |
460 | } | |
461 | ||
462 | __setup("nopfault", nopfault); | |
463 | ||
7dd8fe1f HC |
464 | struct pfault_refbk { |
465 | u16 refdiagc; | |
466 | u16 reffcode; | |
467 | u16 refdwlen; | |
468 | u16 refversn; | |
469 | u64 refgaddr; | |
470 | u64 refselmk; | |
471 | u64 refcmpmk; | |
472 | u64 reserved; | |
473 | } __attribute__ ((packed, aligned(8))); | |
1da177e4 LT |
474 | |
475 | int pfault_init(void) | |
476 | { | |
7dd8fe1f HC |
477 | struct pfault_refbk refbk = { |
478 | .refdiagc = 0x258, | |
479 | .reffcode = 0, | |
480 | .refdwlen = 5, | |
481 | .refversn = 2, | |
482 | .refgaddr = __LC_CURRENT_PID, | |
483 | .refselmk = 1ULL << 48, | |
484 | .refcmpmk = 1ULL << 48, | |
485 | .reserved = __PF_RES_FIELD }; | |
1da177e4 LT |
486 | int rc; |
487 | ||
f32269a0 | 488 | if (pfault_disable) |
1da177e4 | 489 | return -1; |
94c12cc7 MS |
490 | asm volatile( |
491 | " diag %1,%0,0x258\n" | |
492 | "0: j 2f\n" | |
493 | "1: la %0,8\n" | |
1da177e4 | 494 | "2:\n" |
94c12cc7 MS |
495 | EX_TABLE(0b,1b) |
496 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
497 | return rc; |
498 | } | |
499 | ||
500 | void pfault_fini(void) | |
501 | { | |
7dd8fe1f HC |
502 | struct pfault_refbk refbk = { |
503 | .refdiagc = 0x258, | |
504 | .reffcode = 1, | |
505 | .refdwlen = 5, | |
506 | .refversn = 2, | |
507 | }; | |
1da177e4 | 508 | |
f32269a0 | 509 | if (pfault_disable) |
1da177e4 | 510 | return; |
94c12cc7 MS |
511 | asm volatile( |
512 | " diag %0,0,0x258\n" | |
1da177e4 | 513 | "0:\n" |
94c12cc7 MS |
514 | EX_TABLE(0b,0b) |
515 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
516 | } |
517 | ||
f2db2e6c HC |
518 | static DEFINE_SPINLOCK(pfault_lock); |
519 | static LIST_HEAD(pfault_list); | |
520 | ||
fde15c3a | 521 | static void pfault_interrupt(struct ext_code ext_code, |
f6649a7e | 522 | unsigned int param32, unsigned long param64) |
1da177e4 LT |
523 | { |
524 | struct task_struct *tsk; | |
525 | __u16 subcode; | |
f2db2e6c | 526 | pid_t pid; |
1da177e4 LT |
527 | |
528 | /* | |
529 | * Get the external interruption subcode & pfault | |
530 | * initial/completion signal bit. VM stores this | |
531 | * in the 'cpu address' field associated with the | |
532 | * external interrupt. | |
533 | */ | |
fde15c3a | 534 | subcode = ext_code.subcode; |
1da177e4 LT |
535 | if ((subcode & 0xff00) != __SUBCODE_MASK) |
536 | return; | |
420f42ec | 537 | inc_irq_stat(IRQEXT_PFL); |
54c27791 HC |
538 | /* Get the token (= pid of the affected task). */ |
539 | pid = sizeof(void *) == 4 ? param32 : param64; | |
540 | rcu_read_lock(); | |
541 | tsk = find_task_by_pid_ns(pid, &init_pid_ns); | |
542 | if (tsk) | |
543 | get_task_struct(tsk); | |
544 | rcu_read_unlock(); | |
545 | if (!tsk) | |
546 | return; | |
f2db2e6c | 547 | spin_lock(&pfault_lock); |
1da177e4 LT |
548 | if (subcode & 0x0080) { |
549 | /* signal bit is set -> a page has been swapped in by VM */ | |
f2db2e6c | 550 | if (tsk->thread.pfault_wait == 1) { |
1da177e4 LT |
551 | /* Initial interrupt was faster than the completion |
552 | * interrupt. pfault_wait is valid. Set pfault_wait | |
553 | * back to zero and wake up the process. This can | |
554 | * safely be done because the task is still sleeping | |
b6d09449 | 555 | * and can't produce new pfaults. */ |
1da177e4 | 556 | tsk->thread.pfault_wait = 0; |
f2db2e6c | 557 | list_del(&tsk->thread.list); |
1da177e4 | 558 | wake_up_process(tsk); |
d5e50a51 | 559 | put_task_struct(tsk); |
f2db2e6c HC |
560 | } else { |
561 | /* Completion interrupt was faster than initial | |
562 | * interrupt. Set pfault_wait to -1 so the initial | |
fa2fb2f4 HC |
563 | * interrupt doesn't put the task to sleep. |
564 | * If the task is not running, ignore the completion | |
565 | * interrupt since it must be a leftover of a PFAULT | |
566 | * CANCEL operation which didn't remove all pending | |
567 | * completion interrupts. */ | |
568 | if (tsk->state == TASK_RUNNING) | |
569 | tsk->thread.pfault_wait = -1; | |
1da177e4 LT |
570 | } |
571 | } else { | |
572 | /* signal bit not set -> a real page is missing. */ | |
d49f47f8 HC |
573 | if (WARN_ON_ONCE(tsk != current)) |
574 | goto out; | |
d5e50a51 HC |
575 | if (tsk->thread.pfault_wait == 1) { |
576 | /* Already on the list with a reference: put to sleep */ | |
0a16ba78 | 577 | __set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
d5e50a51 HC |
578 | set_tsk_need_resched(tsk); |
579 | } else if (tsk->thread.pfault_wait == -1) { | |
1da177e4 | 580 | /* Completion interrupt was faster than the initial |
f2db2e6c HC |
581 | * interrupt (pfault_wait == -1). Set pfault_wait |
582 | * back to zero and exit. */ | |
1da177e4 | 583 | tsk->thread.pfault_wait = 0; |
f2db2e6c HC |
584 | } else { |
585 | /* Initial interrupt arrived before completion | |
d5e50a51 HC |
586 | * interrupt. Let the task sleep. |
587 | * An extra task reference is needed since a different | |
588 | * cpu may set the task state to TASK_RUNNING again | |
589 | * before the scheduler is reached. */ | |
590 | get_task_struct(tsk); | |
f2db2e6c HC |
591 | tsk->thread.pfault_wait = 1; |
592 | list_add(&tsk->thread.list, &pfault_list); | |
0a16ba78 | 593 | __set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
1da177e4 | 594 | set_tsk_need_resched(tsk); |
f2db2e6c HC |
595 | } |
596 | } | |
d49f47f8 | 597 | out: |
f2db2e6c | 598 | spin_unlock(&pfault_lock); |
54c27791 | 599 | put_task_struct(tsk); |
f2db2e6c HC |
600 | } |
601 | ||
e2741f17 PG |
602 | static int pfault_cpu_notify(struct notifier_block *self, unsigned long action, |
603 | void *hcpu) | |
f2db2e6c HC |
604 | { |
605 | struct thread_struct *thread, *next; | |
606 | struct task_struct *tsk; | |
607 | ||
1c725922 | 608 | switch (action & ~CPU_TASKS_FROZEN) { |
f2db2e6c | 609 | case CPU_DEAD: |
f2db2e6c HC |
610 | spin_lock_irq(&pfault_lock); |
611 | list_for_each_entry_safe(thread, next, &pfault_list, list) { | |
612 | thread->pfault_wait = 0; | |
613 | list_del(&thread->list); | |
614 | tsk = container_of(thread, struct task_struct, thread); | |
615 | wake_up_process(tsk); | |
d5e50a51 | 616 | put_task_struct(tsk); |
f2db2e6c HC |
617 | } |
618 | spin_unlock_irq(&pfault_lock); | |
619 | break; | |
620 | default: | |
621 | break; | |
1da177e4 | 622 | } |
f2db2e6c | 623 | return NOTIFY_OK; |
1da177e4 | 624 | } |
1da177e4 | 625 | |
fb0a9d7e | 626 | static int __init pfault_irq_init(void) |
29b08d2b | 627 | { |
fb0a9d7e | 628 | int rc; |
29b08d2b | 629 | |
fb0a9d7e | 630 | rc = register_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
631 | if (rc) |
632 | goto out_extint; | |
633 | rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; | |
634 | if (rc) | |
635 | goto out_pfault; | |
82003c3e | 636 | irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL); |
7dd8fe1f HC |
637 | hotcpu_notifier(pfault_cpu_notify, 0); |
638 | return 0; | |
29b08d2b | 639 | |
7dd8fe1f | 640 | out_pfault: |
fb0a9d7e | 641 | unregister_external_interrupt(0x2603, pfault_interrupt); |
7dd8fe1f HC |
642 | out_extint: |
643 | pfault_disable = 1; | |
644 | return rc; | |
29b08d2b | 645 | } |
fb0a9d7e HC |
646 | early_initcall(pfault_irq_init); |
647 | ||
7dd8fe1f | 648 | #endif /* CONFIG_PFAULT */ |