arch/s390/mm/fault.c

   1 /*
   2  *  S390 version
   3  *    Copyright IBM Corp. 1999
   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
  11 #include <linux/kernel_stat.h>
  12 #include <linux/perf_event.h>
  13 #include <linux/signal.h>
  14 #include <linux/sched.h>
  15 #include <linux/sched/debug.h>
  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>
  23 #include <linux/compat.h>
  24 #include <linux/smp.h>
  25 #include <linux/kdebug.h>
  26 #include <linux/init.h>
  27 #include <linux/console.h>
  28 #include <linux/extable.h>
  29 #include <linux/hardirq.h>
  30 #include <linux/kprobes.h>
  31 #include <linux/uaccess.h>
  32 #include <linux/hugetlb.h>
  33 #include <asm/asm-offsets.h>
  34 #include <asm/diag.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/gmap.h>
  37 #include <asm/irq.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/facility.h>
  40 #include "../kernel/entry.h"
  41
  42 #define __FAIL_ADDR_MASK -4096L
  43 #define __SUBCODE_MASK 0x0600
  44 #define __PF_RES_FIELD 0x8000000000000000ULL
  45
  46 #define VM_FAULT_BADCONTEXT     0x010000
  47 #define VM_FAULT_BADMAP         0x020000
  48 #define VM_FAULT_BADACCESS      0x040000
  49 #define VM_FAULT_SIGNAL         0x080000
  50 #define VM_FAULT_PFAULT         0x100000
  51
  52 static unsigned long store_indication __read_mostly;
  53
  54 static int __init fault_init(void)
  55 {
  56         if (test_facility(75))
  57                 store_indication = 0xc00;
  58         return 0;
  59 }
  60 early_initcall(fault_init);
  61
  62 static inline int notify_page_fault(struct pt_regs *regs)
  63 {
  64         int ret = 0;
  65
  66         /* kprobe_running() needs smp_processor_id() */
  67         if (kprobes_built_in() && !user_mode(regs)) {
  68                 preempt_disable();
  69                 if (kprobe_running() && kprobe_fault_handler(regs, 14))
  70                         ret = 1;
  71                 preempt_enable();
  72         }
  73         return ret;
  74 }
  75
  76
  77 /*
  78  * Unlock any spinlocks which will prevent us from getting the
  79  * message out.
  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 /*
 101  * Returns the address space associated with the fault.
 102  * Returns 0 for kernel space and 1 for user space.
 103  */
 104 static inline int user_space_fault(struct pt_regs *regs)
 105 {
 106         unsigned long trans_exc_code;
 107
 108         /*
 109          * The lowest two bits of the translation exception
 110          * identification indicate which paging table was used.
 111          */
 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 */
 118                 return current->thread.mm_segment.ar4;
 119         if (current->flags & PF_VCPU)
 120                 return 1;
 121         return 0;
 122 }
 123
 124 static int bad_address(void *p)
 125 {
 126         unsigned long dummy;
 127
 128         return probe_kernel_address((unsigned long *)p, dummy);
 129 }
 130
 131 static void dump_pagetable(unsigned long asce, unsigned long address)
 132 {
 133         unsigned long *table = __va(asce & _ASCE_ORIGIN);
 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;
 168                 pr_cont("S:%016lx ", *table);
 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
 184 static void dump_fault_info(struct pt_regs *regs)
 185 {
 186         unsigned long asce;
 187
 188         pr_alert("Failing address: %016lx TEID: %016lx\n",
 189                  regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
 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
 225 int show_unhandled_signals = 1;
 226
 227 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
 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;
 235         printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
 236                regs->int_code & 0xffff, regs->int_code >> 17);
 237         print_vma_addr(KERN_CONT "in ", regs->psw.addr);
 238         printk(KERN_CONT "\n");
 239         if (is_mm_fault)
 240                 dump_fault_info(regs);
 241         show_regs(regs);
 242 }
 243
 244 /*
 245  * Send SIGSEGV to task.  This is an external routine
 246  * to keep the stack usage of do_page_fault small.
 247  */
 248 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
 249 {
 250         struct siginfo si;
 251
 252         report_user_fault(regs, SIGSEGV, 1);
 253         si.si_signo = SIGSEGV;
 254         si.si_errno = 0;
 255         si.si_code = si_code;
 256         si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
 257         force_sig_info(SIGSEGV, &si, current);
 258 }
 259
 260 static noinline void do_no_context(struct pt_regs *regs)
 261 {
 262         const struct exception_table_entry *fixup;
 263
 264         /* Are we prepared to handle this kernel fault?  */
 265         fixup = search_exception_tables(regs->psw.addr);
 266         if (fixup) {
 267                 regs->psw.addr = extable_fixup(fixup);
 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          */
 275         if (!user_space_fault(regs))
 276                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
 277                        " in virtual kernel address space\n");
 278         else
 279                 printk(KERN_ALERT "Unable to handle kernel paging request"
 280                        " in virtual user address space\n");
 281         dump_fault_info(regs);
 282         die(regs, "Oops");
 283         do_exit(SIGKILL);
 284 }
 285
 286 static noinline void do_low_address(struct pt_regs *regs)
 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'. */
 292                 die (regs, "Low-address protection");
 293                 do_exit(SIGKILL);
 294         }
 295
 296         do_no_context(regs);
 297 }
 298
 299 static noinline void do_sigbus(struct pt_regs *regs)
 300 {
 301         struct task_struct *tsk = current;
 302         struct siginfo si;
 303
 304         /*
 305          * Send a sigbus, regardless of whether we were in kernel
 306          * or user mode.
 307          */
 308         si.si_signo = SIGBUS;
 309         si.si_errno = 0;
 310         si.si_code = BUS_ADRERR;
 311         si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
 312         force_sig_info(SIGBUS, &si, tsk);
 313 }
 314
 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)
 336 {
 337         int si_code;
 338
 339         switch (fault) {
 340         case VM_FAULT_BADACCESS:
 341                 if (access == VM_EXEC && signal_return(regs) == 0)
 342                         break;
 343         case VM_FAULT_BADMAP:
 344                 /* Bad memory access. Check if it is kernel or user space. */
 345                 if (user_mode(regs)) {
 346                         /* User mode accesses just cause a SIGSEGV */
 347                         si_code = (fault == VM_FAULT_BADMAP) ?
 348                                 SEGV_MAPERR : SEGV_ACCERR;
 349                         do_sigsegv(regs, si_code);
 350                         break;
 351                 }
 352         case VM_FAULT_BADCONTEXT:
 353         case VM_FAULT_PFAULT:
 354                 do_no_context(regs);
 355                 break;
 356         case VM_FAULT_SIGNAL:
 357                 if (!user_mode(regs))
 358                         do_no_context(regs);
 359                 break;
 360         default: /* fault & VM_FAULT_ERROR */
 361                 if (fault & VM_FAULT_OOM) {
 362                         if (!user_mode(regs))
 363                                 do_no_context(regs);
 364                         else
 365                                 pagefault_out_of_memory();
 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);
 372                 } else if (fault & VM_FAULT_SIGBUS) {
 373                         /* Kernel mode? Handle exceptions or die */
 374                         if (!user_mode(regs))
 375                                 do_no_context(regs);
 376                         else
 377                                 do_sigbus(regs);
 378                 } else
 379                         BUG();
 380                 break;
 381         }
 382 }
 383
 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  *
 389  * interruption code (int_code):
 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  */
 395 static inline int do_exception(struct pt_regs *regs, int access)
 396 {
 397 #ifdef CONFIG_PGSTE
 398         struct gmap *gmap;
 399 #endif
 400         struct task_struct *tsk;
 401         struct mm_struct *mm;
 402         struct vm_area_struct *vma;
 403         unsigned long trans_exc_code;
 404         unsigned long address;
 405         unsigned int flags;
 406         int fault;
 407
 408         tsk = current;
 409         /*
 410          * The instruction that caused the program check has
 411          * been nullified. Don't signal single step via SIGTRAP.
 412          */
 413         clear_pt_regs_flag(regs, PIF_PER_TRAP);
 414
 415         if (notify_page_fault(regs))
 416                 return 0;
 417
 418         mm = tsk->mm;
 419         trans_exc_code = regs->int_parm_long;
 420
 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          */
 426         fault = VM_FAULT_BADCONTEXT;
 427         if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm))
 428                 goto out;
 429
 430         address = trans_exc_code & __FAIL_ADDR_MASK;
 431         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 432         flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 433         if (user_mode(regs))
 434                 flags |= FAULT_FLAG_USER;
 435         if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
 436                 flags |= FAULT_FLAG_WRITE;
 437         down_read(&mm->mmap_sem);
 438
 439 #ifdef CONFIG_PGSTE
 440         gmap = (current->flags & PF_VCPU) ?
 441                 (struct gmap *) S390_lowcore.gmap : NULL;
 442         if (gmap) {
 443                 current->thread.gmap_addr = address;
 444                 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
 445                 current->thread.gmap_int_code = regs->int_code & 0xffff;
 446                 address = __gmap_translate(gmap, address);
 447                 if (address == -EFAULT) {
 448                         fault = VM_FAULT_BADMAP;
 449                         goto out_up;
 450                 }
 451                 if (gmap->pfault_enabled)
 452                         flags |= FAULT_FLAG_RETRY_NOWAIT;
 453         }
 454 #endif
 455
 456 retry:
 457         fault = VM_FAULT_BADMAP;
 458         vma = find_vma(mm, address);
 459         if (!vma)
 460                 goto out_up;
 461
 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;
 474         if (unlikely(!(vma->vm_flags & access)))
 475                 goto out_up;
 476
 477         if (is_vm_hugetlb_page(vma))
 478                 address &= HPAGE_MASK;
 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          */
 484         fault = handle_mm_fault(vma, address, flags);
 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         }
 490         if (unlikely(fault & VM_FAULT_ERROR))
 491                 goto out_up;
 492
 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++;
 501                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 502                                       regs, address);
 503                 } else {
 504                         tsk->min_flt++;
 505                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 506                                       regs, address);
 507                 }
 508                 if (fault & VM_FAULT_RETRY) {
 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
 518                         /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
 519                          * of starvation. */
 520                         flags &= ~(FAULT_FLAG_ALLOW_RETRY |
 521                                    FAULT_FLAG_RETRY_NOWAIT);
 522                         flags |= FAULT_FLAG_TRIED;
 523                         down_read(&mm->mmap_sem);
 524                         goto retry;
 525                 }
 526         }
 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
 541         fault = 0;
 542 out_up:
 543         up_read(&mm->mmap_sem);
 544 out:
 545         return fault;
 546 }
 547
 548 void do_protection_exception(struct pt_regs *regs)
 549 {
 550         unsigned long trans_exc_code;
 551         int access, fault;
 552
 553         trans_exc_code = regs->int_parm_long;
 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);
 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          */
 566         if (unlikely(!(trans_exc_code & 4))) {
 567                 do_low_address(regs);
 568                 return;
 569         }
 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         }
 579         if (unlikely(fault))
 580                 do_fault_error(regs, access, fault);
 581 }
 582 NOKPROBE_SYMBOL(do_protection_exception);
 583
 584 void do_dat_exception(struct pt_regs *regs)
 585 {
 586         int access, fault;
 587
 588         access = VM_READ | VM_EXEC | VM_WRITE;
 589         fault = do_exception(regs, access);
 590         if (unlikely(fault))
 591                 do_fault_error(regs, access, fault);
 592 }
 593 NOKPROBE_SYMBOL(do_dat_exception);
 594
 595 #ifdef CONFIG_PFAULT
 596 /*
 597  * 'pfault' pseudo page faults routines.
 598  */
 599 static int pfault_disable;
 600
 601 static int __init nopfault(char *str)
 602 {
 603         pfault_disable = 1;
 604         return 1;
 605 }
 606
 607 __setup("nopfault", nopfault);
 608
 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)));
 619
 620 int pfault_init(void)
 621 {
 622         struct pfault_refbk refbk = {
 623                 .refdiagc = 0x258,
 624                 .reffcode = 0,
 625                 .refdwlen = 5,
 626                 .refversn = 2,
 627                 .refgaddr = __LC_LPP,
 628                 .refselmk = 1ULL << 48,
 629                 .refcmpmk = 1ULL << 48,
 630                 .reserved = __PF_RES_FIELD };
 631         int rc;
 632
 633         if (pfault_disable)
 634                 return -1;
 635         diag_stat_inc(DIAG_STAT_X258);
 636         asm volatile(
 637                 "       diag    %1,%0,0x258\n"
 638                 "0:     j       2f\n"
 639                 "1:     la      %0,8\n"
 640                 "2:\n"
 641                 EX_TABLE(0b,1b)
 642                 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
 643         return rc;
 644 }
 645
 646 void pfault_fini(void)
 647 {
 648         struct pfault_refbk refbk = {
 649                 .refdiagc = 0x258,
 650                 .reffcode = 1,
 651                 .refdwlen = 5,
 652                 .refversn = 2,
 653         };
 654
 655         if (pfault_disable)
 656                 return;
 657         diag_stat_inc(DIAG_STAT_X258);
 658         asm volatile(
 659                 "       diag    %0,0,0x258\n"
 660                 "0:     nopr    %%r7\n"
 661                 EX_TABLE(0b,0b)
 662                 : : "a" (&refbk), "m" (refbk) : "cc");
 663 }
 664
 665 static DEFINE_SPINLOCK(pfault_lock);
 666 static LIST_HEAD(pfault_list);
 667
 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  */
 691 static void pfault_interrupt(struct ext_code ext_code,
 692                              unsigned int param32, unsigned long param64)
 693 {
 694         struct task_struct *tsk;
 695         __u16 subcode;
 696         pid_t pid;
 697
 698         /*
 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.
 702          */
 703         subcode = ext_code.subcode;
 704         if ((subcode & 0xff00) != __SUBCODE_MASK)
 705                 return;
 706         inc_irq_stat(IRQEXT_PFL);
 707         /* Get the token (= pid of the affected task). */
 708         pid = param64 & LPP_PFAULT_PID_MASK;
 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;
 716         spin_lock(&pfault_lock);
 717         if (subcode & PF_COMPLETE) {
 718                 /* signal bit is set -> a page has been swapped in by VM */
 719                 if (tsk->thread.pfault_wait == 1) {
 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
 724                          * and can't produce new pfaults. */
 725                         tsk->thread.pfault_wait = 0;
 726                         list_del(&tsk->thread.list);
 727                         wake_up_process(tsk);
 728                         put_task_struct(tsk);
 729                 } else {
 730                         /* Completion interrupt was faster than initial
 731                          * interrupt. Set pfault_wait to -1 so the initial
 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;
 739                 }
 740         } else {
 741                 /* signal bit not set -> a real page is missing. */
 742                 if (WARN_ON_ONCE(tsk != current))
 743                         goto out;
 744                 if (tsk->thread.pfault_wait == 1) {
 745                         /* Already on the list with a reference: put to sleep */
 746                         goto block;
 747                 } else if (tsk->thread.pfault_wait == -1) {
 748                         /* Completion interrupt was faster than the initial
 749                          * interrupt (pfault_wait == -1). Set pfault_wait
 750                          * back to zero and exit. */
 751                         tsk->thread.pfault_wait = 0;
 752                 } else {
 753                         /* Initial interrupt arrived before completion
 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);
 759                         tsk->thread.pfault_wait = 1;
 760                         list_add(&tsk->thread.list, &pfault_list);
 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);
 766                         set_tsk_need_resched(tsk);
 767                         set_preempt_need_resched();
 768                 }
 769         }
 770 out:
 771         spin_unlock(&pfault_lock);
 772         put_task_struct(tsk);
 773 }
 774
 775 static int pfault_cpu_dead(unsigned int cpu)
 776 {
 777         struct thread_struct *thread, *next;
 778         struct task_struct *tsk;
 779
 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);
 787         }
 788         spin_unlock_irq(&pfault_lock);
 789         return 0;
 790 }
 791
 792 static int __init pfault_irq_init(void)
 793 {
 794         int rc;
 795
 796         rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
 797         if (rc)
 798                 goto out_extint;
 799         rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
 800         if (rc)
 801                 goto out_pfault;
 802         irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
 803         cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
 804                                   NULL, pfault_cpu_dead);
 805         return 0;
 806
 807 out_pfault:
 808         unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
 809 out_extint:
 810         pfault_disable = 1;
 811         return rc;
 812 }
 813 early_initcall(pfault_irq_init);
 814
 815 #endif /* CONFIG_PFAULT */