]>
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> | |
b17b0153 | 15 | #include <linux/sched/debug.h> |
1da177e4 LT |
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> | |
dcc096c5 | 28 | #include <linux/extable.h> |
1da177e4 | 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> |
1ec2772e | 34 | #include <asm/diag.h> |
1da177e4 | 35 | #include <asm/pgtable.h> |
1e133ab2 | 36 | #include <asm/gmap.h> |
d7b250e2 | 37 | #include <asm/irq.h> |
6252d702 | 38 | #include <asm/mmu_context.h> |
a0616cde | 39 | #include <asm/facility.h> |
a806170e | 40 | #include "../kernel/entry.h" |
1da177e4 | 41 | |
1da177e4 | 42 | #define __FAIL_ADDR_MASK -4096L |
1da177e4 LT |
43 | #define __SUBCODE_MASK 0x0600 |
44 | #define __PF_RES_FIELD 0x8000000000000000ULL | |
1da177e4 | 45 | |
50d7280d MS |
46 | #define VM_FAULT_BADCONTEXT 0x010000 |
47 | #define VM_FAULT_BADMAP 0x020000 | |
48 | #define VM_FAULT_BADACCESS 0x040000 | |
a4f32bdb | 49 | #define VM_FAULT_SIGNAL 0x080000 |
24eb3a82 | 50 | #define VM_FAULT_PFAULT 0x100000 |
50d7280d | 51 | |
a4f32bdb | 52 | static unsigned long store_indication __read_mostly; |
92f842ea | 53 | |
a4f32bdb | 54 | static int __init fault_init(void) |
92f842ea | 55 | { |
a4f32bdb | 56 | if (test_facility(75)) |
92f842ea | 57 | store_indication = 0xc00; |
a4f32bdb | 58 | return 0; |
92f842ea | 59 | } |
a4f32bdb | 60 | early_initcall(fault_init); |
92f842ea | 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 | */ |
457f2180 | 104 | static inline int user_space_fault(struct pt_regs *regs) |
1da177e4 | 105 | { |
457f2180 HC |
106 | unsigned long trans_exc_code; |
107 | ||
1da177e4 | 108 | /* |
61365e13 MS |
109 | * The lowest two bits of the translation exception |
110 | * identification indicate which paging table was used. | |
1da177e4 | 111 | */ |
457f2180 HC |
112 | trans_exc_code = regs->int_parm_long & 3; |
113 | if (trans_exc_code == 3) /* home space -> kernel */ | |
114 | return 0; | |
115 | if (user_mode(regs)) | |
116 | return 1; | |
117 | if (trans_exc_code == 2) /* secondary space -> set_fs */ | |
61365e13 | 118 | return current->thread.mm_segment.ar4; |
457f2180 HC |
119 | if (current->flags & PF_VCPU) |
120 | return 1; | |
121 | return 0; | |
1da177e4 LT |
122 | } |
123 | ||
3b7df342 HC |
124 | static int bad_address(void *p) |
125 | { | |
126 | unsigned long dummy; | |
127 | ||
128 | return probe_kernel_address((unsigned long *)p, dummy); | |
129 | } | |
130 | ||
3b7df342 HC |
131 | static void dump_pagetable(unsigned long asce, unsigned long address) |
132 | { | |
fe7b2747 | 133 | unsigned long *table = __va(asce & _ASCE_ORIGIN); |
3b7df342 HC |
134 | |
135 | pr_alert("AS:%016lx ", asce); | |
136 | switch (asce & _ASCE_TYPE_MASK) { | |
137 | case _ASCE_TYPE_REGION1: | |
138 | table = table + ((address >> 53) & 0x7ff); | |
139 | if (bad_address(table)) | |
140 | goto bad; | |
141 | pr_cont("R1:%016lx ", *table); | |
142 | if (*table & _REGION_ENTRY_INVALID) | |
143 | goto out; | |
144 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
145 | /* fallthrough */ | |
146 | case _ASCE_TYPE_REGION2: | |
147 | table = table + ((address >> 42) & 0x7ff); | |
148 | if (bad_address(table)) | |
149 | goto bad; | |
150 | pr_cont("R2:%016lx ", *table); | |
151 | if (*table & _REGION_ENTRY_INVALID) | |
152 | goto out; | |
153 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
154 | /* fallthrough */ | |
155 | case _ASCE_TYPE_REGION3: | |
156 | table = table + ((address >> 31) & 0x7ff); | |
157 | if (bad_address(table)) | |
158 | goto bad; | |
159 | pr_cont("R3:%016lx ", *table); | |
160 | if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE)) | |
161 | goto out; | |
162 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
163 | /* fallthrough */ | |
164 | case _ASCE_TYPE_SEGMENT: | |
165 | table = table + ((address >> 20) & 0x7ff); | |
166 | if (bad_address(table)) | |
167 | goto bad; | |
91c0837e | 168 | pr_cont("S:%016lx ", *table); |
3b7df342 HC |
169 | if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE)) |
170 | goto out; | |
171 | table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN); | |
172 | } | |
173 | table = table + ((address >> 12) & 0xff); | |
174 | if (bad_address(table)) | |
175 | goto bad; | |
176 | pr_cont("P:%016lx ", *table); | |
177 | out: | |
178 | pr_cont("\n"); | |
179 | return; | |
180 | bad: | |
181 | pr_cont("BAD\n"); | |
182 | } | |
183 | ||
3b7df342 HC |
184 | static void dump_fault_info(struct pt_regs *regs) |
185 | { | |
186 | unsigned long asce; | |
187 | ||
5d7eccec HC |
188 | pr_alert("Failing address: %016lx TEID: %016lx\n", |
189 | regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long); | |
3b7df342 HC |
190 | pr_alert("Fault in "); |
191 | switch (regs->int_parm_long & 3) { | |
192 | case 3: | |
193 | pr_cont("home space "); | |
194 | break; | |
195 | case 2: | |
196 | pr_cont("secondary space "); | |
197 | break; | |
198 | case 1: | |
199 | pr_cont("access register "); | |
200 | break; | |
201 | case 0: | |
202 | pr_cont("primary space "); | |
203 | break; | |
204 | } | |
205 | pr_cont("mode while using "); | |
206 | if (!user_space_fault(regs)) { | |
207 | asce = S390_lowcore.kernel_asce; | |
208 | pr_cont("kernel "); | |
209 | } | |
210 | #ifdef CONFIG_PGSTE | |
211 | else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) { | |
212 | struct gmap *gmap = (struct gmap *)S390_lowcore.gmap; | |
213 | asce = gmap->asce; | |
214 | pr_cont("gmap "); | |
215 | } | |
216 | #endif | |
217 | else { | |
218 | asce = S390_lowcore.user_asce; | |
219 | pr_cont("user "); | |
220 | } | |
221 | pr_cont("ASCE.\n"); | |
222 | dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK); | |
223 | } | |
224 | ||
5d7eccec HC |
225 | int show_unhandled_signals = 1; |
226 | ||
227 | void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault) | |
ab3c68ee HC |
228 | { |
229 | if ((task_pid_nr(current) > 1) && !show_unhandled_signals) | |
230 | return; | |
231 | if (!unhandled_signal(current, signr)) | |
232 | return; | |
233 | if (!printk_ratelimit()) | |
234 | return; | |
db1177ee | 235 | printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ", |
413d4047 | 236 | regs->int_code & 0xffff, regs->int_code >> 17); |
9cb1ccec | 237 | print_vma_addr(KERN_CONT "in ", regs->psw.addr); |
aa33c8cb | 238 | printk(KERN_CONT "\n"); |
5d7eccec HC |
239 | if (is_mm_fault) |
240 | dump_fault_info(regs); | |
ab3c68ee HC |
241 | show_regs(regs); |
242 | } | |
243 | ||
1da177e4 LT |
244 | /* |
245 | * Send SIGSEGV to task. This is an external routine | |
246 | * to keep the stack usage of do_page_fault small. | |
247 | */ | |
aa33c8cb | 248 | static noinline void do_sigsegv(struct pt_regs *regs, int si_code) |
1da177e4 LT |
249 | { |
250 | struct siginfo si; | |
251 | ||
5d7eccec | 252 | report_user_fault(regs, SIGSEGV, 1); |
1da177e4 | 253 | si.si_signo = SIGSEGV; |
cf0d44d5 | 254 | si.si_errno = 0; |
1da177e4 | 255 | si.si_code = si_code; |
aa33c8cb | 256 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
1da177e4 LT |
257 | force_sig_info(SIGSEGV, &si, current); |
258 | } | |
259 | ||
aa33c8cb | 260 | static noinline void do_no_context(struct pt_regs *regs) |
10c1031f MS |
261 | { |
262 | const struct exception_table_entry *fixup; | |
263 | ||
264 | /* Are we prepared to handle this kernel fault? */ | |
9cb1ccec | 265 | fixup = search_exception_tables(regs->psw.addr); |
10c1031f | 266 | if (fixup) { |
fecc868a | 267 | regs->psw.addr = extable_fixup(fixup); |
10c1031f MS |
268 | return; |
269 | } | |
270 | ||
271 | /* | |
272 | * Oops. The kernel tried to access some bad page. We'll have to | |
273 | * terminate things with extreme prejudice. | |
274 | */ | |
457f2180 | 275 | if (!user_space_fault(regs)) |
10c1031f | 276 | printk(KERN_ALERT "Unable to handle kernel pointer dereference" |
3b7df342 | 277 | " in virtual kernel address space\n"); |
10c1031f MS |
278 | else |
279 | printk(KERN_ALERT "Unable to handle kernel paging request" | |
3b7df342 | 280 | " in virtual user address space\n"); |
3b7df342 | 281 | dump_fault_info(regs); |
aa33c8cb | 282 | die(regs, "Oops"); |
10c1031f MS |
283 | do_exit(SIGKILL); |
284 | } | |
285 | ||
aa33c8cb | 286 | static noinline void do_low_address(struct pt_regs *regs) |
10c1031f MS |
287 | { |
288 | /* Low-address protection hit in kernel mode means | |
289 | NULL pointer write access in kernel mode. */ | |
290 | if (regs->psw.mask & PSW_MASK_PSTATE) { | |
291 | /* Low-address protection hit in user mode 'cannot happen'. */ | |
aa33c8cb | 292 | die (regs, "Low-address protection"); |
10c1031f MS |
293 | do_exit(SIGKILL); |
294 | } | |
295 | ||
aa33c8cb | 296 | do_no_context(regs); |
10c1031f MS |
297 | } |
298 | ||
aa33c8cb | 299 | static noinline void do_sigbus(struct pt_regs *regs) |
10c1031f MS |
300 | { |
301 | struct task_struct *tsk = current; | |
36bf9680 | 302 | struct siginfo si; |
10c1031f | 303 | |
10c1031f MS |
304 | /* |
305 | * Send a sigbus, regardless of whether we were in kernel | |
306 | * or user mode. | |
307 | */ | |
36bf9680 MS |
308 | si.si_signo = SIGBUS; |
309 | si.si_errno = 0; | |
310 | si.si_code = BUS_ADRERR; | |
aa33c8cb | 311 | si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK); |
36bf9680 | 312 | force_sig_info(SIGBUS, &si, tsk); |
10c1031f MS |
313 | } |
314 | ||
57d7f939 MS |
315 | static noinline int signal_return(struct pt_regs *regs) |
316 | { | |
317 | u16 instruction; | |
318 | int rc; | |
319 | ||
320 | rc = __get_user(instruction, (u16 __user *) regs->psw.addr); | |
321 | if (rc) | |
322 | return rc; | |
323 | if (instruction == 0x0a77) { | |
324 | set_pt_regs_flag(regs, PIF_SYSCALL); | |
325 | regs->int_code = 0x00040077; | |
326 | return 0; | |
327 | } else if (instruction == 0x0aad) { | |
328 | set_pt_regs_flag(regs, PIF_SYSCALL); | |
329 | regs->int_code = 0x000400ad; | |
330 | return 0; | |
331 | } | |
332 | return -EACCES; | |
333 | } | |
334 | ||
335 | static noinline void do_fault_error(struct pt_regs *regs, int access, int fault) | |
50d7280d MS |
336 | { |
337 | int si_code; | |
338 | ||
339 | switch (fault) { | |
340 | case VM_FAULT_BADACCESS: | |
57d7f939 MS |
341 | if (access == VM_EXEC && signal_return(regs) == 0) |
342 | break; | |
50d7280d MS |
343 | case VM_FAULT_BADMAP: |
344 | /* Bad memory access. Check if it is kernel or user space. */ | |
7d256175 | 345 | if (user_mode(regs)) { |
50d7280d MS |
346 | /* User mode accesses just cause a SIGSEGV */ |
347 | si_code = (fault == VM_FAULT_BADMAP) ? | |
348 | SEGV_MAPERR : SEGV_ACCERR; | |
aa33c8cb | 349 | do_sigsegv(regs, si_code); |
57d7f939 | 350 | break; |
50d7280d MS |
351 | } |
352 | case VM_FAULT_BADCONTEXT: | |
24eb3a82 | 353 | case VM_FAULT_PFAULT: |
aa33c8cb | 354 | do_no_context(regs); |
50d7280d | 355 | break; |
f2c76e3b HC |
356 | case VM_FAULT_SIGNAL: |
357 | if (!user_mode(regs)) | |
358 | do_no_context(regs); | |
359 | break; | |
50d7280d | 360 | default: /* fault & VM_FAULT_ERROR */ |
99583181 | 361 | if (fault & VM_FAULT_OOM) { |
7d256175 | 362 | if (!user_mode(regs)) |
aa33c8cb | 363 | do_no_context(regs); |
99583181 HC |
364 | else |
365 | pagefault_out_of_memory(); | |
33692f27 LT |
366 | } else if (fault & VM_FAULT_SIGSEGV) { |
367 | /* Kernel mode? Handle exceptions or die */ | |
368 | if (!user_mode(regs)) | |
369 | do_no_context(regs); | |
370 | else | |
371 | do_sigsegv(regs, SEGV_MAPERR); | |
99583181 | 372 | } else if (fault & VM_FAULT_SIGBUS) { |
50d7280d | 373 | /* Kernel mode? Handle exceptions or die */ |
7d256175 | 374 | if (!user_mode(regs)) |
aa33c8cb | 375 | do_no_context(regs); |
36bf9680 | 376 | else |
aa33c8cb | 377 | do_sigbus(regs); |
50d7280d MS |
378 | } else |
379 | BUG(); | |
380 | break; | |
381 | } | |
382 | } | |
383 | ||
1da177e4 LT |
384 | /* |
385 | * This routine handles page faults. It determines the address, | |
386 | * and the problem, and then passes it off to one of the appropriate | |
387 | * routines. | |
388 | * | |
50d7280d | 389 | * interruption code (int_code): |
1da177e4 LT |
390 | * 04 Protection -> Write-Protection (suprression) |
391 | * 10 Segment translation -> Not present (nullification) | |
392 | * 11 Page translation -> Not present (nullification) | |
393 | * 3b Region third trans. -> Not present (nullification) | |
394 | */ | |
aa33c8cb | 395 | static inline int do_exception(struct pt_regs *regs, int access) |
1da177e4 | 396 | { |
24eb3a82 DD |
397 | #ifdef CONFIG_PGSTE |
398 | struct gmap *gmap; | |
399 | #endif | |
10c1031f MS |
400 | struct task_struct *tsk; |
401 | struct mm_struct *mm; | |
402 | struct vm_area_struct *vma; | |
aa33c8cb | 403 | unsigned long trans_exc_code; |
10c1031f | 404 | unsigned long address; |
33ce6140 HC |
405 | unsigned int flags; |
406 | int fault; | |
1da177e4 | 407 | |
39efd4ec MS |
408 | tsk = current; |
409 | /* | |
410 | * The instruction that caused the program check has | |
411 | * been nullified. Don't signal single step via SIGTRAP. | |
412 | */ | |
d3a73acb | 413 | clear_pt_regs_flag(regs, PIF_PER_TRAP); |
39efd4ec | 414 | |
7ecb344a | 415 | if (notify_page_fault(regs)) |
50d7280d | 416 | return 0; |
4ba069b8 | 417 | |
10c1031f | 418 | mm = tsk->mm; |
aa33c8cb | 419 | trans_exc_code = regs->int_parm_long; |
1da177e4 | 420 | |
1da177e4 LT |
421 | /* |
422 | * Verify that the fault happened in user space, that | |
423 | * we are not in an interrupt and that there is a | |
424 | * user context. | |
425 | */ | |
50d7280d | 426 | fault = VM_FAULT_BADCONTEXT; |
70ffdb93 | 427 | if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm)) |
50d7280d | 428 | goto out; |
1da177e4 | 429 | |
61365e13 | 430 | address = trans_exc_code & __FAIL_ADDR_MASK; |
a8b0ca17 | 431 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
f2c76e3b | 432 | flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
759496ba JW |
433 | if (user_mode(regs)) |
434 | flags |= FAULT_FLAG_USER; | |
33ce6140 HC |
435 | if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400) |
436 | flags |= FAULT_FLAG_WRITE; | |
10c1031f | 437 | down_read(&mm->mmap_sem); |
1da177e4 | 438 | |
e5992f2e | 439 | #ifdef CONFIG_PGSTE |
527e30b4 MS |
440 | gmap = (current->flags & PF_VCPU) ? |
441 | (struct gmap *) S390_lowcore.gmap : NULL; | |
24eb3a82 | 442 | if (gmap) { |
527e30b4 | 443 | current->thread.gmap_addr = address; |
4be130a0 | 444 | current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE); |
4a494439 | 445 | current->thread.gmap_int_code = regs->int_code & 0xffff; |
527e30b4 | 446 | address = __gmap_translate(gmap, address); |
e5992f2e MS |
447 | if (address == -EFAULT) { |
448 | fault = VM_FAULT_BADMAP; | |
449 | goto out_up; | |
450 | } | |
24eb3a82 DD |
451 | if (gmap->pfault_enabled) |
452 | flags |= FAULT_FLAG_RETRY_NOWAIT; | |
e5992f2e MS |
453 | } |
454 | #endif | |
455 | ||
456 | retry: | |
50d7280d | 457 | fault = VM_FAULT_BADMAP; |
482b05dd GS |
458 | vma = find_vma(mm, address); |
459 | if (!vma) | |
50d7280d | 460 | goto out_up; |
c1821c2e | 461 | |
50d7280d MS |
462 | if (unlikely(vma->vm_start > address)) { |
463 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
464 | goto out_up; | |
465 | if (expand_stack(vma, address)) | |
466 | goto out_up; | |
467 | } | |
468 | ||
469 | /* | |
470 | * Ok, we have a good vm_area for this memory access, so | |
471 | * we can handle it.. | |
472 | */ | |
473 | fault = VM_FAULT_BADACCESS; | |
1ab947de | 474 | if (unlikely(!(vma->vm_flags & access))) |
50d7280d | 475 | goto out_up; |
1da177e4 | 476 | |
53492b1d GS |
477 | if (is_vm_hugetlb_page(vma)) |
478 | address &= HPAGE_MASK; | |
1da177e4 LT |
479 | /* |
480 | * If for any reason at all we couldn't handle the fault, | |
481 | * make sure we exit gracefully rather than endlessly redo | |
482 | * the fault. | |
483 | */ | |
dcddffd4 | 484 | fault = handle_mm_fault(vma, address, flags); |
f2c76e3b HC |
485 | /* No reason to continue if interrupted by SIGKILL. */ |
486 | if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) { | |
487 | fault = VM_FAULT_SIGNAL; | |
488 | goto out; | |
489 | } | |
50d7280d MS |
490 | if (unlikely(fault & VM_FAULT_ERROR)) |
491 | goto out_up; | |
492 | ||
33ce6140 HC |
493 | /* |
494 | * Major/minor page fault accounting is only done on the | |
495 | * initial attempt. If we go through a retry, it is extremely | |
496 | * likely that the page will be found in page cache at that point. | |
497 | */ | |
498 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
499 | if (fault & VM_FAULT_MAJOR) { | |
500 | tsk->maj_flt++; | |
a8b0ca17 | 501 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
33ce6140 HC |
502 | regs, address); |
503 | } else { | |
504 | tsk->min_flt++; | |
a8b0ca17 | 505 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
33ce6140 HC |
506 | regs, address); |
507 | } | |
508 | if (fault & VM_FAULT_RETRY) { | |
24eb3a82 DD |
509 | #ifdef CONFIG_PGSTE |
510 | if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
511 | /* FAULT_FLAG_RETRY_NOWAIT has been set, | |
512 | * mmap_sem has not been released */ | |
513 | current->thread.gmap_pfault = 1; | |
514 | fault = VM_FAULT_PFAULT; | |
515 | goto out_up; | |
516 | } | |
517 | #endif | |
33ce6140 HC |
518 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk |
519 | * of starvation. */ | |
24eb3a82 DD |
520 | flags &= ~(FAULT_FLAG_ALLOW_RETRY | |
521 | FAULT_FLAG_RETRY_NOWAIT); | |
45cac65b | 522 | flags |= FAULT_FLAG_TRIED; |
e5992f2e | 523 | down_read(&mm->mmap_sem); |
33ce6140 HC |
524 | goto retry; |
525 | } | |
bde69af2 | 526 | } |
527e30b4 MS |
527 | #ifdef CONFIG_PGSTE |
528 | if (gmap) { | |
529 | address = __gmap_link(gmap, current->thread.gmap_addr, | |
530 | address); | |
531 | if (address == -EFAULT) { | |
532 | fault = VM_FAULT_BADMAP; | |
533 | goto out_up; | |
534 | } | |
535 | if (address == -ENOMEM) { | |
536 | fault = VM_FAULT_OOM; | |
537 | goto out_up; | |
538 | } | |
539 | } | |
540 | #endif | |
50d7280d MS |
541 | fault = 0; |
542 | out_up: | |
10c1031f | 543 | up_read(&mm->mmap_sem); |
50d7280d MS |
544 | out: |
545 | return fault; | |
1da177e4 LT |
546 | } |
547 | ||
7a5388de | 548 | void do_protection_exception(struct pt_regs *regs) |
1da177e4 | 549 | { |
aa33c8cb | 550 | unsigned long trans_exc_code; |
57d7f939 | 551 | int access, fault; |
61365e13 | 552 | |
aa33c8cb | 553 | trans_exc_code = regs->int_parm_long; |
f752ac4d MS |
554 | /* |
555 | * Protection exceptions are suppressing, decrement psw address. | |
556 | * The exception to this rule are aborted transactions, for these | |
557 | * the PSW already points to the correct location. | |
558 | */ | |
559 | if (!(regs->int_code & 0x200)) | |
560 | regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16); | |
10c1031f MS |
561 | /* |
562 | * Check for low-address protection. This needs to be treated | |
563 | * as a special case because the translation exception code | |
564 | * field is not guaranteed to contain valid data in this case. | |
565 | */ | |
61365e13 | 566 | if (unlikely(!(trans_exc_code & 4))) { |
aa33c8cb | 567 | do_low_address(regs); |
10c1031f MS |
568 | return; |
569 | } | |
57d7f939 MS |
570 | if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) { |
571 | regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) | | |
572 | (regs->psw.addr & PAGE_MASK); | |
573 | access = VM_EXEC; | |
574 | fault = VM_FAULT_BADACCESS; | |
575 | } else { | |
576 | access = VM_WRITE; | |
577 | fault = do_exception(regs, access); | |
578 | } | |
50d7280d | 579 | if (unlikely(fault)) |
57d7f939 | 580 | do_fault_error(regs, access, fault); |
1da177e4 | 581 | } |
7a5388de | 582 | NOKPROBE_SYMBOL(do_protection_exception); |
1da177e4 | 583 | |
7a5388de | 584 | void do_dat_exception(struct pt_regs *regs) |
1da177e4 | 585 | { |
1ab947de | 586 | int access, fault; |
50d7280d | 587 | |
1ab947de | 588 | access = VM_READ | VM_EXEC | VM_WRITE; |
aa33c8cb | 589 | fault = do_exception(regs, access); |
50d7280d | 590 | if (unlikely(fault)) |
57d7f939 | 591 | do_fault_error(regs, access, fault); |
1da177e4 | 592 | } |
7a5388de | 593 | NOKPROBE_SYMBOL(do_dat_exception); |
1da177e4 | 594 | |
1da177e4 LT |
595 | #ifdef CONFIG_PFAULT |
596 | /* | |
597 | * 'pfault' pseudo page faults routines. | |
598 | */ | |
fb0a9d7e | 599 | static int pfault_disable; |
1da177e4 LT |
600 | |
601 | static int __init nopfault(char *str) | |
602 | { | |
603 | pfault_disable = 1; | |
604 | return 1; | |
605 | } | |
606 | ||
607 | __setup("nopfault", nopfault); | |
608 | ||
7dd8fe1f HC |
609 | struct pfault_refbk { |
610 | u16 refdiagc; | |
611 | u16 reffcode; | |
612 | u16 refdwlen; | |
613 | u16 refversn; | |
614 | u64 refgaddr; | |
615 | u64 refselmk; | |
616 | u64 refcmpmk; | |
617 | u64 reserved; | |
618 | } __attribute__ ((packed, aligned(8))); | |
1da177e4 LT |
619 | |
620 | int pfault_init(void) | |
621 | { | |
7dd8fe1f HC |
622 | struct pfault_refbk refbk = { |
623 | .refdiagc = 0x258, | |
624 | .reffcode = 0, | |
625 | .refdwlen = 5, | |
626 | .refversn = 2, | |
e22cf8ca | 627 | .refgaddr = __LC_LPP, |
7dd8fe1f HC |
628 | .refselmk = 1ULL << 48, |
629 | .refcmpmk = 1ULL << 48, | |
630 | .reserved = __PF_RES_FIELD }; | |
1da177e4 LT |
631 | int rc; |
632 | ||
f32269a0 | 633 | if (pfault_disable) |
1da177e4 | 634 | return -1; |
1ec2772e | 635 | diag_stat_inc(DIAG_STAT_X258); |
94c12cc7 MS |
636 | asm volatile( |
637 | " diag %1,%0,0x258\n" | |
638 | "0: j 2f\n" | |
639 | "1: la %0,8\n" | |
1da177e4 | 640 | "2:\n" |
94c12cc7 MS |
641 | EX_TABLE(0b,1b) |
642 | : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
643 | return rc; |
644 | } | |
645 | ||
646 | void pfault_fini(void) | |
647 | { | |
7dd8fe1f HC |
648 | struct pfault_refbk refbk = { |
649 | .refdiagc = 0x258, | |
650 | .reffcode = 1, | |
651 | .refdwlen = 5, | |
652 | .refversn = 2, | |
653 | }; | |
1da177e4 | 654 | |
f32269a0 | 655 | if (pfault_disable) |
1da177e4 | 656 | return; |
1ec2772e | 657 | diag_stat_inc(DIAG_STAT_X258); |
94c12cc7 MS |
658 | asm volatile( |
659 | " diag %0,0,0x258\n" | |
6c22c986 | 660 | "0: nopr %%r7\n" |
94c12cc7 MS |
661 | EX_TABLE(0b,0b) |
662 | : : "a" (&refbk), "m" (refbk) : "cc"); | |
1da177e4 LT |
663 | } |
664 | ||
f2db2e6c HC |
665 | static DEFINE_SPINLOCK(pfault_lock); |
666 | static LIST_HEAD(pfault_list); | |
667 | ||
0227f7c4 PZ |
668 | #define PF_COMPLETE 0x0080 |
669 | ||
670 | /* | |
671 | * The mechanism of our pfault code: if Linux is running as guest, runs a user | |
672 | * space process and the user space process accesses a page that the host has | |
673 | * paged out we get a pfault interrupt. | |
674 | * | |
675 | * This allows us, within the guest, to schedule a different process. Without | |
676 | * this mechanism the host would have to suspend the whole virtual cpu until | |
677 | * the page has been paged in. | |
678 | * | |
679 | * So when we get such an interrupt then we set the state of the current task | |
680 | * to uninterruptible and also set the need_resched flag. Both happens within | |
681 | * interrupt context(!). If we later on want to return to user space we | |
682 | * recognize the need_resched flag and then call schedule(). It's not very | |
683 | * obvious how this works... | |
684 | * | |
685 | * Of course we have a lot of additional fun with the completion interrupt (-> | |
686 | * host signals that a page of a process has been paged in and the process can | |
687 | * continue to run). This interrupt can arrive on any cpu and, since we have | |
688 | * virtual cpus, actually appear before the interrupt that signals that a page | |
689 | * is missing. | |
690 | */ | |
fde15c3a | 691 | static void pfault_interrupt(struct ext_code ext_code, |
f6649a7e | 692 | unsigned int param32, unsigned long param64) |
1da177e4 LT |
693 | { |
694 | struct task_struct *tsk; | |
695 | __u16 subcode; | |
f2db2e6c | 696 | pid_t pid; |
1da177e4 LT |
697 | |
698 | /* | |
0227f7c4 PZ |
699 | * Get the external interruption subcode & pfault initial/completion |
700 | * signal bit. VM stores this in the 'cpu address' field associated | |
701 | * with the external interrupt. | |
1da177e4 | 702 | */ |
fde15c3a | 703 | subcode = ext_code.subcode; |
1da177e4 LT |
704 | if ((subcode & 0xff00) != __SUBCODE_MASK) |
705 | return; | |
420f42ec | 706 | inc_irq_stat(IRQEXT_PFL); |
54c27791 | 707 | /* Get the token (= pid of the affected task). */ |
e22cf8ca | 708 | pid = param64 & LPP_PFAULT_PID_MASK; |
54c27791 HC |
709 | rcu_read_lock(); |
710 | tsk = find_task_by_pid_ns(pid, &init_pid_ns); | |
711 | if (tsk) | |
712 | get_task_struct(tsk); | |
713 | rcu_read_unlock(); | |
714 | if (!tsk) | |
715 | return; | |
f2db2e6c | 716 | spin_lock(&pfault_lock); |
0227f7c4 | 717 | if (subcode & PF_COMPLETE) { |
1da177e4 | 718 | /* signal bit is set -> a page has been swapped in by VM */ |
f2db2e6c | 719 | if (tsk->thread.pfault_wait == 1) { |
1da177e4 LT |
720 | /* Initial interrupt was faster than the completion |
721 | * interrupt. pfault_wait is valid. Set pfault_wait | |
722 | * back to zero and wake up the process. This can | |
723 | * safely be done because the task is still sleeping | |
b6d09449 | 724 | * and can't produce new pfaults. */ |
1da177e4 | 725 | tsk->thread.pfault_wait = 0; |
f2db2e6c | 726 | list_del(&tsk->thread.list); |
1da177e4 | 727 | wake_up_process(tsk); |
d5e50a51 | 728 | put_task_struct(tsk); |
f2db2e6c HC |
729 | } else { |
730 | /* Completion interrupt was faster than initial | |
731 | * interrupt. Set pfault_wait to -1 so the initial | |
fa2fb2f4 HC |
732 | * interrupt doesn't put the task to sleep. |
733 | * If the task is not running, ignore the completion | |
734 | * interrupt since it must be a leftover of a PFAULT | |
735 | * CANCEL operation which didn't remove all pending | |
736 | * completion interrupts. */ | |
737 | if (tsk->state == TASK_RUNNING) | |
738 | tsk->thread.pfault_wait = -1; | |
1da177e4 LT |
739 | } |
740 | } else { | |
741 | /* signal bit not set -> a real page is missing. */ | |
d49f47f8 HC |
742 | if (WARN_ON_ONCE(tsk != current)) |
743 | goto out; | |
d5e50a51 HC |
744 | if (tsk->thread.pfault_wait == 1) { |
745 | /* Already on the list with a reference: put to sleep */ | |
0227f7c4 | 746 | goto block; |
d5e50a51 | 747 | } else if (tsk->thread.pfault_wait == -1) { |
1da177e4 | 748 | /* Completion interrupt was faster than the initial |
f2db2e6c HC |
749 | * interrupt (pfault_wait == -1). Set pfault_wait |
750 | * back to zero and exit. */ | |
1da177e4 | 751 | tsk->thread.pfault_wait = 0; |
f2db2e6c HC |
752 | } else { |
753 | /* Initial interrupt arrived before completion | |
d5e50a51 HC |
754 | * interrupt. Let the task sleep. |
755 | * An extra task reference is needed since a different | |
756 | * cpu may set the task state to TASK_RUNNING again | |
757 | * before the scheduler is reached. */ | |
758 | get_task_struct(tsk); | |
f2db2e6c HC |
759 | tsk->thread.pfault_wait = 1; |
760 | list_add(&tsk->thread.list, &pfault_list); | |
0227f7c4 PZ |
761 | block: |
762 | /* Since this must be a userspace fault, there | |
763 | * is no kernel task state to trample. Rely on the | |
764 | * return to userspace schedule() to block. */ | |
765 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1da177e4 | 766 | set_tsk_need_resched(tsk); |
c360192b | 767 | set_preempt_need_resched(); |
f2db2e6c HC |
768 | } |
769 | } | |
d49f47f8 | 770 | out: |
f2db2e6c | 771 | spin_unlock(&pfault_lock); |
54c27791 | 772 | put_task_struct(tsk); |
f2db2e6c HC |
773 | } |
774 | ||
84c9ceef | 775 | static int pfault_cpu_dead(unsigned int cpu) |
f2db2e6c HC |
776 | { |
777 | struct thread_struct *thread, *next; | |
778 | struct task_struct *tsk; | |
779 | ||
84c9ceef SAS |
780 | spin_lock_irq(&pfault_lock); |
781 | list_for_each_entry_safe(thread, next, &pfault_list, list) { | |
782 | thread->pfault_wait = 0; | |
783 | list_del(&thread->list); | |
784 | tsk = container_of(thread, struct task_struct, thread); | |
785 | wake_up_process(tsk); | |
786 | put_task_struct(tsk); | |
1da177e4 | 787 | } |
84c9ceef SAS |
788 | spin_unlock_irq(&pfault_lock); |
789 | return 0; | |
1da177e4 | 790 | } |
1da177e4 | 791 | |
fb0a9d7e | 792 | static int __init pfault_irq_init(void) |
29b08d2b | 793 | { |
fb0a9d7e | 794 | int rc; |
29b08d2b | 795 | |
1dad093b | 796 | rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt); |
7dd8fe1f HC |
797 | if (rc) |
798 | goto out_extint; | |
799 | rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; | |
800 | if (rc) | |
801 | goto out_pfault; | |
82003c3e | 802 | irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL); |
84c9ceef SAS |
803 | cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead", |
804 | NULL, pfault_cpu_dead); | |
7dd8fe1f | 805 | return 0; |
29b08d2b | 806 | |
7dd8fe1f | 807 | out_pfault: |
1dad093b | 808 | unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt); |
7dd8fe1f HC |
809 | out_extint: |
810 | pfault_disable = 1; | |
811 | return rc; | |
29b08d2b | 812 | } |
fb0a9d7e HC |
813 | early_initcall(pfault_irq_init); |
814 | ||
7dd8fe1f | 815 | #endif /* CONFIG_PFAULT */ |