]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blame - arch/s390/mm/fault.c
s390/ftrace,kprobes: allow to patch first instruction
[mirror_ubuntu-focal-kernel.git] / arch / s390 / mm / fault.c
CommitLineData
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
24eb3a82 53#define VM_FAULT_PFAULT 0x100000
50d7280d 54
a4f32bdb 55static unsigned long store_indication __read_mostly;
92f842ea 56
a4f32bdb
HC
57#ifdef CONFIG_64BIT
58static int __init fault_init(void)
92f842ea 59{
a4f32bdb 60 if (test_facility(75))
92f842ea 61 store_indication = 0xc00;
a4f32bdb 62 return 0;
92f842ea 63}
a4f32bdb
HC
64early_initcall(fault_init);
65#endif
92f842ea 66
7ecb344a 67static inline int notify_page_fault(struct pt_regs *regs)
10c1031f 68{
33464e3b
CH
69 int ret = 0;
70
71 /* kprobe_running() needs smp_processor_id() */
22e0a046 72 if (kprobes_built_in() && !user_mode(regs)) {
33464e3b
CH
73 preempt_disable();
74 if (kprobe_running() && kprobe_fault_handler(regs, 14))
75 ret = 1;
76 preempt_enable();
77 }
33464e3b 78 return ret;
4ba069b8 79}
4ba069b8 80
1da177e4
LT
81
82/*
83 * Unlock any spinlocks which will prevent us from getting the
cefc8be8 84 * message out.
1da177e4
LT
85 */
86void bust_spinlocks(int yes)
87{
88 if (yes) {
89 oops_in_progress = 1;
90 } else {
91 int loglevel_save = console_loglevel;
92 console_unblank();
93 oops_in_progress = 0;
94 /*
95 * OK, the message is on the console. Now we call printk()
96 * without oops_in_progress set so that printk will give klogd
97 * a poke. Hold onto your hats...
98 */
99 console_loglevel = 15;
100 printk(" ");
101 console_loglevel = loglevel_save;
102 }
103}
104
105/*
482b05dd 106 * Returns the address space associated with the fault.
61365e13 107 * Returns 0 for kernel space and 1 for user space.
1da177e4 108 */
457f2180 109static inline int user_space_fault(struct pt_regs *regs)
1da177e4 110{
457f2180
HC
111 unsigned long trans_exc_code;
112
1da177e4 113 /*
61365e13
MS
114 * The lowest two bits of the translation exception
115 * identification indicate which paging table was used.
1da177e4 116 */
457f2180
HC
117 trans_exc_code = regs->int_parm_long & 3;
118 if (trans_exc_code == 3) /* home space -> kernel */
119 return 0;
120 if (user_mode(regs))
121 return 1;
122 if (trans_exc_code == 2) /* secondary space -> set_fs */
61365e13 123 return current->thread.mm_segment.ar4;
457f2180
HC
124 if (current->flags & PF_VCPU)
125 return 1;
126 return 0;
1da177e4
LT
127}
128
3b7df342
HC
129static int bad_address(void *p)
130{
131 unsigned long dummy;
132
133 return probe_kernel_address((unsigned long *)p, dummy);
134}
135
136#ifdef CONFIG_64BIT
137static void dump_pagetable(unsigned long asce, unsigned long address)
138{
139 unsigned long *table = __va(asce & PAGE_MASK);
140
141 pr_alert("AS:%016lx ", asce);
142 switch (asce & _ASCE_TYPE_MASK) {
143 case _ASCE_TYPE_REGION1:
144 table = table + ((address >> 53) & 0x7ff);
145 if (bad_address(table))
146 goto bad;
147 pr_cont("R1:%016lx ", *table);
148 if (*table & _REGION_ENTRY_INVALID)
149 goto out;
150 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
151 /* fallthrough */
152 case _ASCE_TYPE_REGION2:
153 table = table + ((address >> 42) & 0x7ff);
154 if (bad_address(table))
155 goto bad;
156 pr_cont("R2:%016lx ", *table);
157 if (*table & _REGION_ENTRY_INVALID)
158 goto out;
159 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
160 /* fallthrough */
161 case _ASCE_TYPE_REGION3:
162 table = table + ((address >> 31) & 0x7ff);
163 if (bad_address(table))
164 goto bad;
165 pr_cont("R3:%016lx ", *table);
166 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
167 goto out;
168 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
169 /* fallthrough */
170 case _ASCE_TYPE_SEGMENT:
171 table = table + ((address >> 20) & 0x7ff);
172 if (bad_address(table))
173 goto bad;
174 pr_cont(KERN_CONT "S:%016lx ", *table);
175 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
176 goto out;
177 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
178 }
179 table = table + ((address >> 12) & 0xff);
180 if (bad_address(table))
181 goto bad;
182 pr_cont("P:%016lx ", *table);
183out:
184 pr_cont("\n");
185 return;
186bad:
187 pr_cont("BAD\n");
188}
189
190#else /* CONFIG_64BIT */
191
192static void dump_pagetable(unsigned long asce, unsigned long address)
193{
194 unsigned long *table = __va(asce & PAGE_MASK);
195
196 pr_alert("AS:%08lx ", asce);
197 table = table + ((address >> 20) & 0x7ff);
198 if (bad_address(table))
199 goto bad;
200 pr_cont("S:%08lx ", *table);
201 if (*table & _SEGMENT_ENTRY_INVALID)
202 goto out;
203 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
204 table = table + ((address >> 12) & 0xff);
205 if (bad_address(table))
206 goto bad;
207 pr_cont("P:%08lx ", *table);
208out:
209 pr_cont("\n");
210 return;
211bad:
212 pr_cont("BAD\n");
213}
214
215#endif /* CONFIG_64BIT */
216
217static void dump_fault_info(struct pt_regs *regs)
218{
219 unsigned long asce;
220
221 pr_alert("Fault in ");
222 switch (regs->int_parm_long & 3) {
223 case 3:
224 pr_cont("home space ");
225 break;
226 case 2:
227 pr_cont("secondary space ");
228 break;
229 case 1:
230 pr_cont("access register ");
231 break;
232 case 0:
233 pr_cont("primary space ");
234 break;
235 }
236 pr_cont("mode while using ");
237 if (!user_space_fault(regs)) {
238 asce = S390_lowcore.kernel_asce;
239 pr_cont("kernel ");
240 }
241#ifdef CONFIG_PGSTE
242 else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
243 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
244 asce = gmap->asce;
245 pr_cont("gmap ");
246 }
247#endif
248 else {
249 asce = S390_lowcore.user_asce;
250 pr_cont("user ");
251 }
252 pr_cont("ASCE.\n");
253 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
254}
255
aa33c8cb 256static inline void report_user_fault(struct pt_regs *regs, long signr)
ab3c68ee
HC
257{
258 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
259 return;
260 if (!unhandled_signal(current, signr))
261 return;
262 if (!printk_ratelimit())
263 return;
aa33c8cb
MS
264 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
265 regs->int_code);
ab3c68ee 266 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
aa33c8cb 267 printk(KERN_CONT "\n");
3b7df342
HC
268 printk(KERN_ALERT "failing address: %016lx TEID: %016lx\n",
269 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
270 dump_fault_info(regs);
ab3c68ee
HC
271 show_regs(regs);
272}
273
1da177e4
LT
274/*
275 * Send SIGSEGV to task. This is an external routine
276 * to keep the stack usage of do_page_fault small.
277 */
aa33c8cb 278static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
1da177e4
LT
279{
280 struct siginfo si;
281
aa33c8cb 282 report_user_fault(regs, SIGSEGV);
1da177e4
LT
283 si.si_signo = SIGSEGV;
284 si.si_code = si_code;
aa33c8cb 285 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
1da177e4
LT
286 force_sig_info(SIGSEGV, &si, current);
287}
288
aa33c8cb 289static noinline void do_no_context(struct pt_regs *regs)
10c1031f
MS
290{
291 const struct exception_table_entry *fixup;
61365e13 292 unsigned long address;
10c1031f
MS
293
294 /* Are we prepared to handle this kernel fault? */
50d7280d 295 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
10c1031f 296 if (fixup) {
eb608fb3 297 regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE;
10c1031f
MS
298 return;
299 }
300
301 /*
302 * Oops. The kernel tried to access some bad page. We'll have to
303 * terminate things with extreme prejudice.
304 */
aa33c8cb 305 address = regs->int_parm_long & __FAIL_ADDR_MASK;
457f2180 306 if (!user_space_fault(regs))
10c1031f 307 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
3b7df342 308 " in virtual kernel address space\n");
10c1031f
MS
309 else
310 printk(KERN_ALERT "Unable to handle kernel paging request"
3b7df342
HC
311 " in virtual user address space\n");
312 printk(KERN_ALERT "failing address: %016lx TEID: %016lx\n",
313 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
314 dump_fault_info(regs);
aa33c8cb 315 die(regs, "Oops");
10c1031f
MS
316 do_exit(SIGKILL);
317}
318
aa33c8cb 319static noinline void do_low_address(struct pt_regs *regs)
10c1031f
MS
320{
321 /* Low-address protection hit in kernel mode means
322 NULL pointer write access in kernel mode. */
323 if (regs->psw.mask & PSW_MASK_PSTATE) {
324 /* Low-address protection hit in user mode 'cannot happen'. */
aa33c8cb 325 die (regs, "Low-address protection");
10c1031f
MS
326 do_exit(SIGKILL);
327 }
328
aa33c8cb 329 do_no_context(regs);
10c1031f
MS
330}
331
aa33c8cb 332static noinline void do_sigbus(struct pt_regs *regs)
10c1031f
MS
333{
334 struct task_struct *tsk = current;
36bf9680 335 struct siginfo si;
10c1031f 336
10c1031f
MS
337 /*
338 * Send a sigbus, regardless of whether we were in kernel
339 * or user mode.
340 */
36bf9680
MS
341 si.si_signo = SIGBUS;
342 si.si_errno = 0;
343 si.si_code = BUS_ADRERR;
aa33c8cb 344 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
36bf9680 345 force_sig_info(SIGBUS, &si, tsk);
10c1031f
MS
346}
347
aa33c8cb 348static noinline void do_fault_error(struct pt_regs *regs, int fault)
50d7280d
MS
349{
350 int si_code;
351
352 switch (fault) {
353 case VM_FAULT_BADACCESS:
50d7280d
MS
354 case VM_FAULT_BADMAP:
355 /* Bad memory access. Check if it is kernel or user space. */
7d256175 356 if (user_mode(regs)) {
50d7280d
MS
357 /* User mode accesses just cause a SIGSEGV */
358 si_code = (fault == VM_FAULT_BADMAP) ?
359 SEGV_MAPERR : SEGV_ACCERR;
aa33c8cb 360 do_sigsegv(regs, si_code);
50d7280d
MS
361 return;
362 }
363 case VM_FAULT_BADCONTEXT:
24eb3a82 364 case VM_FAULT_PFAULT:
aa33c8cb 365 do_no_context(regs);
50d7280d 366 break;
f2c76e3b
HC
367 case VM_FAULT_SIGNAL:
368 if (!user_mode(regs))
369 do_no_context(regs);
370 break;
50d7280d 371 default: /* fault & VM_FAULT_ERROR */
99583181 372 if (fault & VM_FAULT_OOM) {
7d256175 373 if (!user_mode(regs))
aa33c8cb 374 do_no_context(regs);
99583181
HC
375 else
376 pagefault_out_of_memory();
377 } else if (fault & VM_FAULT_SIGBUS) {
50d7280d 378 /* Kernel mode? Handle exceptions or die */
7d256175 379 if (!user_mode(regs))
aa33c8cb 380 do_no_context(regs);
36bf9680 381 else
aa33c8cb 382 do_sigbus(regs);
50d7280d
MS
383 } else
384 BUG();
385 break;
386 }
387}
388
1da177e4
LT
389/*
390 * This routine handles page faults. It determines the address,
391 * and the problem, and then passes it off to one of the appropriate
392 * routines.
393 *
50d7280d 394 * interruption code (int_code):
1da177e4
LT
395 * 04 Protection -> Write-Protection (suprression)
396 * 10 Segment translation -> Not present (nullification)
397 * 11 Page translation -> Not present (nullification)
398 * 3b Region third trans. -> Not present (nullification)
399 */
aa33c8cb 400static inline int do_exception(struct pt_regs *regs, int access)
1da177e4 401{
24eb3a82
DD
402#ifdef CONFIG_PGSTE
403 struct gmap *gmap;
404#endif
10c1031f
MS
405 struct task_struct *tsk;
406 struct mm_struct *mm;
407 struct vm_area_struct *vma;
aa33c8cb 408 unsigned long trans_exc_code;
10c1031f 409 unsigned long address;
33ce6140
HC
410 unsigned int flags;
411 int fault;
1da177e4 412
39efd4ec
MS
413 tsk = current;
414 /*
415 * The instruction that caused the program check has
416 * been nullified. Don't signal single step via SIGTRAP.
417 */
d3a73acb 418 clear_pt_regs_flag(regs, PIF_PER_TRAP);
39efd4ec 419
7ecb344a 420 if (notify_page_fault(regs))
50d7280d 421 return 0;
4ba069b8 422
10c1031f 423 mm = tsk->mm;
aa33c8cb 424 trans_exc_code = regs->int_parm_long;
1da177e4 425
1da177e4
LT
426 /*
427 * Verify that the fault happened in user space, that
428 * we are not in an interrupt and that there is a
429 * user context.
430 */
50d7280d 431 fault = VM_FAULT_BADCONTEXT;
457f2180 432 if (unlikely(!user_space_fault(regs) || in_atomic() || !mm))
50d7280d 433 goto out;
1da177e4 434
61365e13 435 address = trans_exc_code & __FAIL_ADDR_MASK;
a8b0ca17 436 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
f2c76e3b 437 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
759496ba
JW
438 if (user_mode(regs))
439 flags |= FAULT_FLAG_USER;
33ce6140
HC
440 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
441 flags |= FAULT_FLAG_WRITE;
10c1031f 442 down_read(&mm->mmap_sem);
1da177e4 443
e5992f2e 444#ifdef CONFIG_PGSTE
527e30b4
MS
445 gmap = (current->flags & PF_VCPU) ?
446 (struct gmap *) S390_lowcore.gmap : NULL;
24eb3a82 447 if (gmap) {
527e30b4
MS
448 current->thread.gmap_addr = address;
449 address = __gmap_translate(gmap, address);
e5992f2e
MS
450 if (address == -EFAULT) {
451 fault = VM_FAULT_BADMAP;
452 goto out_up;
453 }
24eb3a82
DD
454 if (gmap->pfault_enabled)
455 flags |= FAULT_FLAG_RETRY_NOWAIT;
e5992f2e
MS
456 }
457#endif
458
459retry:
50d7280d 460 fault = VM_FAULT_BADMAP;
482b05dd
GS
461 vma = find_vma(mm, address);
462 if (!vma)
50d7280d 463 goto out_up;
c1821c2e 464
50d7280d
MS
465 if (unlikely(vma->vm_start > address)) {
466 if (!(vma->vm_flags & VM_GROWSDOWN))
467 goto out_up;
468 if (expand_stack(vma, address))
469 goto out_up;
470 }
471
472 /*
473 * Ok, we have a good vm_area for this memory access, so
474 * we can handle it..
475 */
476 fault = VM_FAULT_BADACCESS;
1ab947de 477 if (unlikely(!(vma->vm_flags & access)))
50d7280d 478 goto out_up;
1da177e4 479
53492b1d
GS
480 if (is_vm_hugetlb_page(vma))
481 address &= HPAGE_MASK;
1da177e4
LT
482 /*
483 * If for any reason at all we couldn't handle the fault,
484 * make sure we exit gracefully rather than endlessly redo
485 * the fault.
486 */
33ce6140 487 fault = handle_mm_fault(mm, vma, address, flags);
f2c76e3b
HC
488 /* No reason to continue if interrupted by SIGKILL. */
489 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
490 fault = VM_FAULT_SIGNAL;
491 goto out;
492 }
50d7280d
MS
493 if (unlikely(fault & VM_FAULT_ERROR))
494 goto out_up;
495
33ce6140
HC
496 /*
497 * Major/minor page fault accounting is only done on the
498 * initial attempt. If we go through a retry, it is extremely
499 * likely that the page will be found in page cache at that point.
500 */
501 if (flags & FAULT_FLAG_ALLOW_RETRY) {
502 if (fault & VM_FAULT_MAJOR) {
503 tsk->maj_flt++;
a8b0ca17 504 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
33ce6140
HC
505 regs, address);
506 } else {
507 tsk->min_flt++;
a8b0ca17 508 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
33ce6140
HC
509 regs, address);
510 }
511 if (fault & VM_FAULT_RETRY) {
24eb3a82
DD
512#ifdef CONFIG_PGSTE
513 if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
514 /* FAULT_FLAG_RETRY_NOWAIT has been set,
515 * mmap_sem has not been released */
516 current->thread.gmap_pfault = 1;
517 fault = VM_FAULT_PFAULT;
518 goto out_up;
519 }
520#endif
33ce6140
HC
521 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
522 * of starvation. */
24eb3a82
DD
523 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
524 FAULT_FLAG_RETRY_NOWAIT);
45cac65b 525 flags |= FAULT_FLAG_TRIED;
e5992f2e 526 down_read(&mm->mmap_sem);
33ce6140
HC
527 goto retry;
528 }
bde69af2 529 }
527e30b4
MS
530#ifdef CONFIG_PGSTE
531 if (gmap) {
532 address = __gmap_link(gmap, current->thread.gmap_addr,
533 address);
534 if (address == -EFAULT) {
535 fault = VM_FAULT_BADMAP;
536 goto out_up;
537 }
538 if (address == -ENOMEM) {
539 fault = VM_FAULT_OOM;
540 goto out_up;
541 }
542 }
543#endif
50d7280d
MS
544 fault = 0;
545out_up:
10c1031f 546 up_read(&mm->mmap_sem);
50d7280d
MS
547out:
548 return fault;
1da177e4
LT
549}
550
aa33c8cb 551void __kprobes do_protection_exception(struct pt_regs *regs)
1da177e4 552{
aa33c8cb 553 unsigned long trans_exc_code;
50d7280d 554 int fault;
61365e13 555
aa33c8cb 556 trans_exc_code = regs->int_parm_long;
f752ac4d
MS
557 /*
558 * Protection exceptions are suppressing, decrement psw address.
559 * The exception to this rule are aborted transactions, for these
560 * the PSW already points to the correct location.
561 */
562 if (!(regs->int_code & 0x200))
563 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
10c1031f
MS
564 /*
565 * Check for low-address protection. This needs to be treated
566 * as a special case because the translation exception code
567 * field is not guaranteed to contain valid data in this case.
568 */
61365e13 569 if (unlikely(!(trans_exc_code & 4))) {
aa33c8cb 570 do_low_address(regs);
10c1031f
MS
571 return;
572 }
aa33c8cb 573 fault = do_exception(regs, VM_WRITE);
50d7280d 574 if (unlikely(fault))
aa33c8cb 575 do_fault_error(regs, fault);
1da177e4
LT
576}
577
aa33c8cb 578void __kprobes do_dat_exception(struct pt_regs *regs)
1da177e4 579{
1ab947de 580 int access, fault;
50d7280d 581
1ab947de 582 access = VM_READ | VM_EXEC | VM_WRITE;
aa33c8cb 583 fault = do_exception(regs, access);
50d7280d 584 if (unlikely(fault))
aa33c8cb 585 do_fault_error(regs, fault);
1da177e4
LT
586}
587
1da177e4
LT
588#ifdef CONFIG_PFAULT
589/*
590 * 'pfault' pseudo page faults routines.
591 */
fb0a9d7e 592static int pfault_disable;
1da177e4
LT
593
594static int __init nopfault(char *str)
595{
596 pfault_disable = 1;
597 return 1;
598}
599
600__setup("nopfault", nopfault);
601
7dd8fe1f
HC
602struct pfault_refbk {
603 u16 refdiagc;
604 u16 reffcode;
605 u16 refdwlen;
606 u16 refversn;
607 u64 refgaddr;
608 u64 refselmk;
609 u64 refcmpmk;
610 u64 reserved;
611} __attribute__ ((packed, aligned(8)));
1da177e4
LT
612
613int pfault_init(void)
614{
7dd8fe1f
HC
615 struct pfault_refbk refbk = {
616 .refdiagc = 0x258,
617 .reffcode = 0,
618 .refdwlen = 5,
619 .refversn = 2,
620 .refgaddr = __LC_CURRENT_PID,
621 .refselmk = 1ULL << 48,
622 .refcmpmk = 1ULL << 48,
623 .reserved = __PF_RES_FIELD };
1da177e4
LT
624 int rc;
625
f32269a0 626 if (pfault_disable)
1da177e4 627 return -1;
94c12cc7
MS
628 asm volatile(
629 " diag %1,%0,0x258\n"
630 "0: j 2f\n"
631 "1: la %0,8\n"
1da177e4 632 "2:\n"
94c12cc7
MS
633 EX_TABLE(0b,1b)
634 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
1da177e4
LT
635 return rc;
636}
637
638void pfault_fini(void)
639{
7dd8fe1f
HC
640 struct pfault_refbk refbk = {
641 .refdiagc = 0x258,
642 .reffcode = 1,
643 .refdwlen = 5,
644 .refversn = 2,
645 };
1da177e4 646
f32269a0 647 if (pfault_disable)
1da177e4 648 return;
94c12cc7
MS
649 asm volatile(
650 " diag %0,0,0x258\n"
1da177e4 651 "0:\n"
94c12cc7
MS
652 EX_TABLE(0b,0b)
653 : : "a" (&refbk), "m" (refbk) : "cc");
1da177e4
LT
654}
655
f2db2e6c
HC
656static DEFINE_SPINLOCK(pfault_lock);
657static LIST_HEAD(pfault_list);
658
fde15c3a 659static void pfault_interrupt(struct ext_code ext_code,
f6649a7e 660 unsigned int param32, unsigned long param64)
1da177e4
LT
661{
662 struct task_struct *tsk;
663 __u16 subcode;
f2db2e6c 664 pid_t pid;
1da177e4
LT
665
666 /*
667 * Get the external interruption subcode & pfault
668 * initial/completion signal bit. VM stores this
669 * in the 'cpu address' field associated with the
670 * external interrupt.
671 */
fde15c3a 672 subcode = ext_code.subcode;
1da177e4
LT
673 if ((subcode & 0xff00) != __SUBCODE_MASK)
674 return;
420f42ec 675 inc_irq_stat(IRQEXT_PFL);
54c27791
HC
676 /* Get the token (= pid of the affected task). */
677 pid = sizeof(void *) == 4 ? param32 : param64;
678 rcu_read_lock();
679 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
680 if (tsk)
681 get_task_struct(tsk);
682 rcu_read_unlock();
683 if (!tsk)
684 return;
f2db2e6c 685 spin_lock(&pfault_lock);
1da177e4
LT
686 if (subcode & 0x0080) {
687 /* signal bit is set -> a page has been swapped in by VM */
f2db2e6c 688 if (tsk->thread.pfault_wait == 1) {
1da177e4
LT
689 /* Initial interrupt was faster than the completion
690 * interrupt. pfault_wait is valid. Set pfault_wait
691 * back to zero and wake up the process. This can
692 * safely be done because the task is still sleeping
b6d09449 693 * and can't produce new pfaults. */
1da177e4 694 tsk->thread.pfault_wait = 0;
f2db2e6c 695 list_del(&tsk->thread.list);
1da177e4 696 wake_up_process(tsk);
d5e50a51 697 put_task_struct(tsk);
f2db2e6c
HC
698 } else {
699 /* Completion interrupt was faster than initial
700 * interrupt. Set pfault_wait to -1 so the initial
fa2fb2f4
HC
701 * interrupt doesn't put the task to sleep.
702 * If the task is not running, ignore the completion
703 * interrupt since it must be a leftover of a PFAULT
704 * CANCEL operation which didn't remove all pending
705 * completion interrupts. */
706 if (tsk->state == TASK_RUNNING)
707 tsk->thread.pfault_wait = -1;
1da177e4
LT
708 }
709 } else {
710 /* signal bit not set -> a real page is missing. */
d49f47f8
HC
711 if (WARN_ON_ONCE(tsk != current))
712 goto out;
d5e50a51
HC
713 if (tsk->thread.pfault_wait == 1) {
714 /* Already on the list with a reference: put to sleep */
0a16ba78 715 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
d5e50a51
HC
716 set_tsk_need_resched(tsk);
717 } else if (tsk->thread.pfault_wait == -1) {
1da177e4 718 /* Completion interrupt was faster than the initial
f2db2e6c
HC
719 * interrupt (pfault_wait == -1). Set pfault_wait
720 * back to zero and exit. */
1da177e4 721 tsk->thread.pfault_wait = 0;
f2db2e6c
HC
722 } else {
723 /* Initial interrupt arrived before completion
d5e50a51
HC
724 * interrupt. Let the task sleep.
725 * An extra task reference is needed since a different
726 * cpu may set the task state to TASK_RUNNING again
727 * before the scheduler is reached. */
728 get_task_struct(tsk);
f2db2e6c
HC
729 tsk->thread.pfault_wait = 1;
730 list_add(&tsk->thread.list, &pfault_list);
0a16ba78 731 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
1da177e4 732 set_tsk_need_resched(tsk);
f2db2e6c
HC
733 }
734 }
d49f47f8 735out:
f2db2e6c 736 spin_unlock(&pfault_lock);
54c27791 737 put_task_struct(tsk);
f2db2e6c
HC
738}
739
e2741f17
PG
740static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
741 void *hcpu)
f2db2e6c
HC
742{
743 struct thread_struct *thread, *next;
744 struct task_struct *tsk;
745
1c725922 746 switch (action & ~CPU_TASKS_FROZEN) {
f2db2e6c 747 case CPU_DEAD:
f2db2e6c
HC
748 spin_lock_irq(&pfault_lock);
749 list_for_each_entry_safe(thread, next, &pfault_list, list) {
750 thread->pfault_wait = 0;
751 list_del(&thread->list);
752 tsk = container_of(thread, struct task_struct, thread);
753 wake_up_process(tsk);
d5e50a51 754 put_task_struct(tsk);
f2db2e6c
HC
755 }
756 spin_unlock_irq(&pfault_lock);
757 break;
758 default:
759 break;
1da177e4 760 }
f2db2e6c 761 return NOTIFY_OK;
1da177e4 762}
1da177e4 763
fb0a9d7e 764static int __init pfault_irq_init(void)
29b08d2b 765{
fb0a9d7e 766 int rc;
29b08d2b 767
1dad093b 768 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
7dd8fe1f
HC
769 if (rc)
770 goto out_extint;
771 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
772 if (rc)
773 goto out_pfault;
82003c3e 774 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
7dd8fe1f
HC
775 hotcpu_notifier(pfault_cpu_notify, 0);
776 return 0;
29b08d2b 777
7dd8fe1f 778out_pfault:
1dad093b 779 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
7dd8fe1f
HC
780out_extint:
781 pfault_disable = 1;
782 return rc;
29b08d2b 783}
fb0a9d7e
HC
784early_initcall(pfault_irq_init);
785
7dd8fe1f 786#endif /* CONFIG_PFAULT */