arch/s390/mm/fault.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  S390 version
   4  *    Copyright IBM Corp. 1999
   5  *    Author(s): Hartmut Penner (hp@de.ibm.com)
   6  *               Ulrich Weigand (uweigand@de.ibm.com)
   7  *
   8  *  Derived from "arch/i386/mm/fault.c"
   9  *    Copyright (C) 1995  Linus Torvalds
  10  */
  11
  12 #include <linux/kernel_stat.h>
  13 #include <linux/perf_event.h>
  14 #include <linux/signal.h>
  15 #include <linux/sched.h>
  16 #include <linux/sched/debug.h>
  17 #include <linux/kernel.h>
  18 #include <linux/errno.h>
  19 #include <linux/string.h>
  20 #include <linux/types.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/mman.h>
  23 #include <linux/mm.h>
  24 #include <linux/compat.h>
  25 #include <linux/smp.h>
  26 #include <linux/kdebug.h>
  27 #include <linux/init.h>
  28 #include <linux/console.h>
  29 #include <linux/extable.h>
  30 #include <linux/hardirq.h>
  31 #include <linux/kprobes.h>
  32 #include <linux/uaccess.h>
  33 #include <linux/hugetlb.h>
  34 #include <asm/asm-offsets.h>
  35 #include <asm/diag.h>
  36 #include <asm/gmap.h>
  37 #include <asm/irq.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/facility.h>
  40 #include <asm/uv.h>
  41 #include "../kernel/entry.h"
  42
  43 #define __FAIL_ADDR_MASK -4096L
  44 #define __SUBCODE_MASK 0x0600
  45 #define __PF_RES_FIELD 0x8000000000000000ULL
  46
  47 #define VM_FAULT_BADCONTEXT     ((__force vm_fault_t) 0x010000)
  48 #define VM_FAULT_BADMAP         ((__force vm_fault_t) 0x020000)
  49 #define VM_FAULT_BADACCESS      ((__force vm_fault_t) 0x040000)
  50 #define VM_FAULT_SIGNAL         ((__force vm_fault_t) 0x080000)
  51 #define VM_FAULT_PFAULT         ((__force vm_fault_t) 0x100000)
  52
  53 enum fault_type {
  54         KERNEL_FAULT,
  55         USER_FAULT,
  56         GMAP_FAULT,
  57 };
  58
  59 static unsigned long store_indication __read_mostly;
  60
  61 static int __init fault_init(void)
  62 {
  63         if (test_facility(75))
  64                 store_indication = 0xc00;
  65         return 0;
  66 }
  67 early_initcall(fault_init);
  68
  69 /*
  70  * Find out which address space caused the exception.
  71  */
  72 static enum fault_type get_fault_type(struct pt_regs *regs)
  73 {
  74         unsigned long trans_exc_code;
  75
  76         trans_exc_code = regs->int_parm_long & 3;
  77         if (likely(trans_exc_code == 0)) {
  78                 /* primary space exception */
  79                 if (user_mode(regs))
  80                         return USER_FAULT;
  81                 if (!IS_ENABLED(CONFIG_PGSTE))
  82                         return KERNEL_FAULT;
  83                 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
  84                         return GMAP_FAULT;
  85                 return KERNEL_FAULT;
  86         }
  87         if (trans_exc_code == 2)
  88                 return USER_FAULT;
  89         if (trans_exc_code == 1) {
  90                 /* access register mode, not used in the kernel */
  91                 return USER_FAULT;
  92         }
  93         /* home space exception -> access via kernel ASCE */
  94         return KERNEL_FAULT;
  95 }
  96
  97 static int bad_address(void *p)
  98 {
  99         unsigned long dummy;
 100
 101         return get_kernel_nofault(dummy, (unsigned long *)p);
 102 }
 103
 104 static void dump_pagetable(unsigned long asce, unsigned long address)
 105 {
 106         unsigned long *table = __va(asce & _ASCE_ORIGIN);
 107
 108         pr_alert("AS:%016lx ", asce);
 109         switch (asce & _ASCE_TYPE_MASK) {
 110         case _ASCE_TYPE_REGION1:
 111                 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
 112                 if (bad_address(table))
 113                         goto bad;
 114                 pr_cont("R1:%016lx ", *table);
 115                 if (*table & _REGION_ENTRY_INVALID)
 116                         goto out;
 117                 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 118                 fallthrough;
 119         case _ASCE_TYPE_REGION2:
 120                 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
 121                 if (bad_address(table))
 122                         goto bad;
 123                 pr_cont("R2:%016lx ", *table);
 124                 if (*table & _REGION_ENTRY_INVALID)
 125                         goto out;
 126                 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 127                 fallthrough;
 128         case _ASCE_TYPE_REGION3:
 129                 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
 130                 if (bad_address(table))
 131                         goto bad;
 132                 pr_cont("R3:%016lx ", *table);
 133                 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
 134                         goto out;
 135                 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 136                 fallthrough;
 137         case _ASCE_TYPE_SEGMENT:
 138                 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 139                 if (bad_address(table))
 140                         goto bad;
 141                 pr_cont("S:%016lx ", *table);
 142                 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
 143                         goto out;
 144                 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
 145         }
 146         table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
 147         if (bad_address(table))
 148                 goto bad;
 149         pr_cont("P:%016lx ", *table);
 150 out:
 151         pr_cont("\n");
 152         return;
 153 bad:
 154         pr_cont("BAD\n");
 155 }
 156
 157 static void dump_fault_info(struct pt_regs *regs)
 158 {
 159         unsigned long asce;
 160
 161         pr_alert("Failing address: %016lx TEID: %016lx\n",
 162                  regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
 163         pr_alert("Fault in ");
 164         switch (regs->int_parm_long & 3) {
 165         case 3:
 166                 pr_cont("home space ");
 167                 break;
 168         case 2:
 169                 pr_cont("secondary space ");
 170                 break;
 171         case 1:
 172                 pr_cont("access register ");
 173                 break;
 174         case 0:
 175                 pr_cont("primary space ");
 176                 break;
 177         }
 178         pr_cont("mode while using ");
 179         switch (get_fault_type(regs)) {
 180         case USER_FAULT:
 181                 asce = S390_lowcore.user_asce;
 182                 pr_cont("user ");
 183                 break;
 184         case GMAP_FAULT:
 185                 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
 186                 pr_cont("gmap ");
 187                 break;
 188         case KERNEL_FAULT:
 189                 asce = S390_lowcore.kernel_asce;
 190                 pr_cont("kernel ");
 191                 break;
 192         default:
 193                 unreachable();
 194         }
 195         pr_cont("ASCE.\n");
 196         dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
 197 }
 198
 199 int show_unhandled_signals = 1;
 200
 201 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
 202 {
 203         if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
 204                 return;
 205         if (!unhandled_signal(current, signr))
 206                 return;
 207         if (!printk_ratelimit())
 208                 return;
 209         printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
 210                regs->int_code & 0xffff, regs->int_code >> 17);
 211         print_vma_addr(KERN_CONT "in ", regs->psw.addr);
 212         printk(KERN_CONT "\n");
 213         if (is_mm_fault)
 214                 dump_fault_info(regs);
 215         show_regs(regs);
 216 }
 217
 218 /*
 219  * Send SIGSEGV to task.  This is an external routine
 220  * to keep the stack usage of do_page_fault small.
 221  */
 222 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
 223 {
 224         report_user_fault(regs, SIGSEGV, 1);
 225         force_sig_fault(SIGSEGV, si_code,
 226                         (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
 227 }
 228
 229 const struct exception_table_entry *s390_search_extables(unsigned long addr)
 230 {
 231         const struct exception_table_entry *fixup;
 232
 233         fixup = search_extable(__start_dma_ex_table,
 234                                __stop_dma_ex_table - __start_dma_ex_table,
 235                                addr);
 236         if (!fixup)
 237                 fixup = search_exception_tables(addr);
 238         return fixup;
 239 }
 240
 241 static noinline void do_no_context(struct pt_regs *regs)
 242 {
 243         const struct exception_table_entry *fixup;
 244
 245         /* Are we prepared to handle this kernel fault?  */
 246         fixup = s390_search_extables(regs->psw.addr);
 247         if (fixup && ex_handle(fixup, regs))
 248                 return;
 249
 250         /*
 251          * Oops. The kernel tried to access some bad page. We'll have to
 252          * terminate things with extreme prejudice.
 253          */
 254         if (get_fault_type(regs) == KERNEL_FAULT)
 255                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
 256                        " in virtual kernel address space\n");
 257         else
 258                 printk(KERN_ALERT "Unable to handle kernel paging request"
 259                        " in virtual user address space\n");
 260         dump_fault_info(regs);
 261         die(regs, "Oops");
 262         do_exit(SIGKILL);
 263 }
 264
 265 static noinline void do_low_address(struct pt_regs *regs)
 266 {
 267         /* Low-address protection hit in kernel mode means
 268            NULL pointer write access in kernel mode.  */
 269         if (regs->psw.mask & PSW_MASK_PSTATE) {
 270                 /* Low-address protection hit in user mode 'cannot happen'. */
 271                 die (regs, "Low-address protection");
 272                 do_exit(SIGKILL);
 273         }
 274
 275         do_no_context(regs);
 276 }
 277
 278 static noinline void do_sigbus(struct pt_regs *regs)
 279 {
 280         /*
 281          * Send a sigbus, regardless of whether we were in kernel
 282          * or user mode.
 283          */
 284         force_sig_fault(SIGBUS, BUS_ADRERR,
 285                         (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
 286 }
 287
 288 static noinline int signal_return(struct pt_regs *regs)
 289 {
 290         u16 instruction;
 291         int rc;
 292
 293         rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
 294         if (rc)
 295                 return rc;
 296         if (instruction == 0x0a77) {
 297                 set_pt_regs_flag(regs, PIF_SYSCALL);
 298                 regs->int_code = 0x00040077;
 299                 return 0;
 300         } else if (instruction == 0x0aad) {
 301                 set_pt_regs_flag(regs, PIF_SYSCALL);
 302                 regs->int_code = 0x000400ad;
 303                 return 0;
 304         }
 305         return -EACCES;
 306 }
 307
 308 static noinline void do_fault_error(struct pt_regs *regs, int access,
 309                                         vm_fault_t fault)
 310 {
 311         int si_code;
 312
 313         switch (fault) {
 314         case VM_FAULT_BADACCESS:
 315                 if (access == VM_EXEC && signal_return(regs) == 0)
 316                         break;
 317                 fallthrough;
 318         case VM_FAULT_BADMAP:
 319                 /* Bad memory access. Check if it is kernel or user space. */
 320                 if (user_mode(regs)) {
 321                         /* User mode accesses just cause a SIGSEGV */
 322                         si_code = (fault == VM_FAULT_BADMAP) ?
 323                                 SEGV_MAPERR : SEGV_ACCERR;
 324                         do_sigsegv(regs, si_code);
 325                         break;
 326                 }
 327                 fallthrough;
 328         case VM_FAULT_BADCONTEXT:
 329         case VM_FAULT_PFAULT:
 330                 do_no_context(regs);
 331                 break;
 332         case VM_FAULT_SIGNAL:
 333                 if (!user_mode(regs))
 334                         do_no_context(regs);
 335                 break;
 336         default: /* fault & VM_FAULT_ERROR */
 337                 if (fault & VM_FAULT_OOM) {
 338                         if (!user_mode(regs))
 339                                 do_no_context(regs);
 340                         else
 341                                 pagefault_out_of_memory();
 342                 } else if (fault & VM_FAULT_SIGSEGV) {
 343                         /* Kernel mode? Handle exceptions or die */
 344                         if (!user_mode(regs))
 345                                 do_no_context(regs);
 346                         else
 347                                 do_sigsegv(regs, SEGV_MAPERR);
 348                 } else if (fault & VM_FAULT_SIGBUS) {
 349                         /* Kernel mode? Handle exceptions or die */
 350                         if (!user_mode(regs))
 351                                 do_no_context(regs);
 352                         else
 353                                 do_sigbus(regs);
 354                 } else
 355                         BUG();
 356                 break;
 357         }
 358 }
 359
 360 /*
 361  * This routine handles page faults.  It determines the address,
 362  * and the problem, and then passes it off to one of the appropriate
 363  * routines.
 364  *
 365  * interruption code (int_code):
 366  *   04       Protection           ->  Write-Protection  (suppression)
 367  *   10       Segment translation  ->  Not present       (nullification)
 368  *   11       Page translation     ->  Not present       (nullification)
 369  *   3b       Region third trans.  ->  Not present       (nullification)
 370  */
 371 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
 372 {
 373         struct gmap *gmap;
 374         struct task_struct *tsk;
 375         struct mm_struct *mm;
 376         struct vm_area_struct *vma;
 377         enum fault_type type;
 378         unsigned long trans_exc_code;
 379         unsigned long address;
 380         unsigned int flags;
 381         vm_fault_t fault;
 382
 383         tsk = current;
 384         /*
 385          * The instruction that caused the program check has
 386          * been nullified. Don't signal single step via SIGTRAP.
 387          */
 388         clear_pt_regs_flag(regs, PIF_PER_TRAP);
 389
 390         if (kprobe_page_fault(regs, 14))
 391                 return 0;
 392
 393         mm = tsk->mm;
 394         trans_exc_code = regs->int_parm_long;
 395
 396         /*
 397          * Verify that the fault happened in user space, that
 398          * we are not in an interrupt and that there is a
 399          * user context.
 400          */
 401         fault = VM_FAULT_BADCONTEXT;
 402         type = get_fault_type(regs);
 403         switch (type) {
 404         case KERNEL_FAULT:
 405                 goto out;
 406         case USER_FAULT:
 407         case GMAP_FAULT:
 408                 if (faulthandler_disabled() || !mm)
 409                         goto out;
 410                 break;
 411         }
 412
 413         address = trans_exc_code & __FAIL_ADDR_MASK;
 414         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 415         flags = FAULT_FLAG_DEFAULT;
 416         if (user_mode(regs))
 417                 flags |= FAULT_FLAG_USER;
 418         if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
 419                 flags |= FAULT_FLAG_WRITE;
 420         mmap_read_lock(mm);
 421
 422         gmap = NULL;
 423         if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
 424                 gmap = (struct gmap *) S390_lowcore.gmap;
 425                 current->thread.gmap_addr = address;
 426                 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
 427                 current->thread.gmap_int_code = regs->int_code & 0xffff;
 428                 address = __gmap_translate(gmap, address);
 429                 if (address == -EFAULT) {
 430                         fault = VM_FAULT_BADMAP;
 431                         goto out_up;
 432                 }
 433                 if (gmap->pfault_enabled)
 434                         flags |= FAULT_FLAG_RETRY_NOWAIT;
 435         }
 436
 437 retry:
 438         fault = VM_FAULT_BADMAP;
 439         vma = find_vma(mm, address);
 440         if (!vma)
 441                 goto out_up;
 442
 443         if (unlikely(vma->vm_start > address)) {
 444                 if (!(vma->vm_flags & VM_GROWSDOWN))
 445                         goto out_up;
 446                 if (expand_stack(vma, address))
 447                         goto out_up;
 448         }
 449
 450         /*
 451          * Ok, we have a good vm_area for this memory access, so
 452          * we can handle it..
 453          */
 454         fault = VM_FAULT_BADACCESS;
 455         if (unlikely(!(vma->vm_flags & access)))
 456                 goto out_up;
 457
 458         if (is_vm_hugetlb_page(vma))
 459                 address &= HPAGE_MASK;
 460         /*
 461          * If for any reason at all we couldn't handle the fault,
 462          * make sure we exit gracefully rather than endlessly redo
 463          * the fault.
 464          */
 465         fault = handle_mm_fault(vma, address, flags, regs);
 466         if (fault_signal_pending(fault, regs)) {
 467                 fault = VM_FAULT_SIGNAL;
 468                 if (flags & FAULT_FLAG_RETRY_NOWAIT)
 469                         goto out_up;
 470                 goto out;
 471         }
 472         if (unlikely(fault & VM_FAULT_ERROR))
 473                 goto out_up;
 474
 475         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 476                 if (fault & VM_FAULT_RETRY) {
 477                         if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
 478                             (flags & FAULT_FLAG_RETRY_NOWAIT)) {
 479                                 /* FAULT_FLAG_RETRY_NOWAIT has been set,
 480                                  * mmap_lock has not been released */
 481                                 current->thread.gmap_pfault = 1;
 482                                 fault = VM_FAULT_PFAULT;
 483                                 goto out_up;
 484                         }
 485                         flags &= ~FAULT_FLAG_RETRY_NOWAIT;
 486                         flags |= FAULT_FLAG_TRIED;
 487                         mmap_read_lock(mm);
 488                         goto retry;
 489                 }
 490         }
 491         if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
 492                 address =  __gmap_link(gmap, current->thread.gmap_addr,
 493                                        address);
 494                 if (address == -EFAULT) {
 495                         fault = VM_FAULT_BADMAP;
 496                         goto out_up;
 497                 }
 498                 if (address == -ENOMEM) {
 499                         fault = VM_FAULT_OOM;
 500                         goto out_up;
 501                 }
 502         }
 503         fault = 0;
 504 out_up:
 505         mmap_read_unlock(mm);
 506 out:
 507         return fault;
 508 }
 509
 510 void do_protection_exception(struct pt_regs *regs)
 511 {
 512         unsigned long trans_exc_code;
 513         int access;
 514         vm_fault_t fault;
 515
 516         trans_exc_code = regs->int_parm_long;
 517         /*
 518          * Protection exceptions are suppressing, decrement psw address.
 519          * The exception to this rule are aborted transactions, for these
 520          * the PSW already points to the correct location.
 521          */
 522         if (!(regs->int_code & 0x200))
 523                 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
 524         /*
 525          * Check for low-address protection.  This needs to be treated
 526          * as a special case because the translation exception code
 527          * field is not guaranteed to contain valid data in this case.
 528          */
 529         if (unlikely(!(trans_exc_code & 4))) {
 530                 do_low_address(regs);
 531                 return;
 532         }
 533         if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
 534                 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
 535                                         (regs->psw.addr & PAGE_MASK);
 536                 access = VM_EXEC;
 537                 fault = VM_FAULT_BADACCESS;
 538         } else {
 539                 access = VM_WRITE;
 540                 fault = do_exception(regs, access);
 541         }
 542         if (unlikely(fault))
 543                 do_fault_error(regs, access, fault);
 544 }
 545 NOKPROBE_SYMBOL(do_protection_exception);
 546
 547 void do_dat_exception(struct pt_regs *regs)
 548 {
 549         int access;
 550         vm_fault_t fault;
 551
 552         access = VM_ACCESS_FLAGS;
 553         fault = do_exception(regs, access);
 554         if (unlikely(fault))
 555                 do_fault_error(regs, access, fault);
 556 }
 557 NOKPROBE_SYMBOL(do_dat_exception);
 558
 559 #ifdef CONFIG_PFAULT
 560 /*
 561  * 'pfault' pseudo page faults routines.
 562  */
 563 static int pfault_disable;
 564
 565 static int __init nopfault(char *str)
 566 {
 567         pfault_disable = 1;
 568         return 1;
 569 }
 570
 571 __setup("nopfault", nopfault);
 572
 573 struct pfault_refbk {
 574         u16 refdiagc;
 575         u16 reffcode;
 576         u16 refdwlen;
 577         u16 refversn;
 578         u64 refgaddr;
 579         u64 refselmk;
 580         u64 refcmpmk;
 581         u64 reserved;
 582 } __attribute__ ((packed, aligned(8)));
 583
 584 static struct pfault_refbk pfault_init_refbk = {
 585         .refdiagc = 0x258,
 586         .reffcode = 0,
 587         .refdwlen = 5,
 588         .refversn = 2,
 589         .refgaddr = __LC_LPP,
 590         .refselmk = 1ULL << 48,
 591         .refcmpmk = 1ULL << 48,
 592         .reserved = __PF_RES_FIELD
 593 };
 594
 595 int pfault_init(void)
 596 {
 597         int rc;
 598
 599         if (pfault_disable)
 600                 return -1;
 601         diag_stat_inc(DIAG_STAT_X258);
 602         asm volatile(
 603                 "       diag    %1,%0,0x258\n"
 604                 "0:     j       2f\n"
 605                 "1:     la      %0,8\n"
 606                 "2:\n"
 607                 EX_TABLE(0b,1b)
 608                 : "=d" (rc)
 609                 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
 610         return rc;
 611 }
 612
 613 static struct pfault_refbk pfault_fini_refbk = {
 614         .refdiagc = 0x258,
 615         .reffcode = 1,
 616         .refdwlen = 5,
 617         .refversn = 2,
 618 };
 619
 620 void pfault_fini(void)
 621 {
 622
 623         if (pfault_disable)
 624                 return;
 625         diag_stat_inc(DIAG_STAT_X258);
 626         asm volatile(
 627                 "       diag    %0,0,0x258\n"
 628                 "0:     nopr    %%r7\n"
 629                 EX_TABLE(0b,0b)
 630                 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
 631 }
 632
 633 static DEFINE_SPINLOCK(pfault_lock);
 634 static LIST_HEAD(pfault_list);
 635
 636 #define PF_COMPLETE     0x0080
 637
 638 /*
 639  * The mechanism of our pfault code: if Linux is running as guest, runs a user
 640  * space process and the user space process accesses a page that the host has
 641  * paged out we get a pfault interrupt.
 642  *
 643  * This allows us, within the guest, to schedule a different process. Without
 644  * this mechanism the host would have to suspend the whole virtual cpu until
 645  * the page has been paged in.
 646  *
 647  * So when we get such an interrupt then we set the state of the current task
 648  * to uninterruptible and also set the need_resched flag. Both happens within
 649  * interrupt context(!). If we later on want to return to user space we
 650  * recognize the need_resched flag and then call schedule().  It's not very
 651  * obvious how this works...
 652  *
 653  * Of course we have a lot of additional fun with the completion interrupt (->
 654  * host signals that a page of a process has been paged in and the process can
 655  * continue to run). This interrupt can arrive on any cpu and, since we have
 656  * virtual cpus, actually appear before the interrupt that signals that a page
 657  * is missing.
 658  */
 659 static void pfault_interrupt(struct ext_code ext_code,
 660                              unsigned int param32, unsigned long param64)
 661 {
 662         struct task_struct *tsk;
 663         __u16 subcode;
 664         pid_t pid;
 665
 666         /*
 667          * Get the external interruption subcode & pfault initial/completion
 668          * signal bit. VM stores this in the 'cpu address' field associated
 669          * with the external interrupt.
 670          */
 671         subcode = ext_code.subcode;
 672         if ((subcode & 0xff00) != __SUBCODE_MASK)
 673                 return;
 674         inc_irq_stat(IRQEXT_PFL);
 675         /* Get the token (= pid of the affected task). */
 676         pid = param64 & LPP_PID_MASK;
 677         rcu_read_lock();
 678         tsk = find_task_by_pid_ns(pid, &init_pid_ns);
 679         if (tsk)
 680                 get_task_struct(tsk);
 681         rcu_read_unlock();
 682         if (!tsk)
 683                 return;
 684         spin_lock(&pfault_lock);
 685         if (subcode & PF_COMPLETE) {
 686                 /* signal bit is set -> a page has been swapped in by VM */
 687                 if (tsk->thread.pfault_wait == 1) {
 688                         /* Initial interrupt was faster than the completion
 689                          * interrupt. pfault_wait is valid. Set pfault_wait
 690                          * back to zero and wake up the process. This can
 691                          * safely be done because the task is still sleeping
 692                          * and can't produce new pfaults. */
 693                         tsk->thread.pfault_wait = 0;
 694                         list_del(&tsk->thread.list);
 695                         wake_up_process(tsk);
 696                         put_task_struct(tsk);
 697                 } else {
 698                         /* Completion interrupt was faster than initial
 699                          * interrupt. Set pfault_wait to -1 so the initial
 700                          * interrupt doesn't put the task to sleep.
 701                          * If the task is not running, ignore the completion
 702                          * interrupt since it must be a leftover of a PFAULT
 703                          * CANCEL operation which didn't remove all pending
 704                          * completion interrupts. */
 705                         if (tsk->state == TASK_RUNNING)
 706                                 tsk->thread.pfault_wait = -1;
 707                 }
 708         } else {
 709                 /* signal bit not set -> a real page is missing. */
 710                 if (WARN_ON_ONCE(tsk != current))
 711                         goto out;
 712                 if (tsk->thread.pfault_wait == 1) {
 713                         /* Already on the list with a reference: put to sleep */
 714                         goto block;
 715                 } else if (tsk->thread.pfault_wait == -1) {
 716                         /* Completion interrupt was faster than the initial
 717                          * interrupt (pfault_wait == -1). Set pfault_wait
 718                          * back to zero and exit. */
 719                         tsk->thread.pfault_wait = 0;
 720                 } else {
 721                         /* Initial interrupt arrived before completion
 722                          * interrupt. Let the task sleep.
 723                          * An extra task reference is needed since a different
 724                          * cpu may set the task state to TASK_RUNNING again
 725                          * before the scheduler is reached. */
 726                         get_task_struct(tsk);
 727                         tsk->thread.pfault_wait = 1;
 728                         list_add(&tsk->thread.list, &pfault_list);
 729 block:
 730                         /* Since this must be a userspace fault, there
 731                          * is no kernel task state to trample. Rely on the
 732                          * return to userspace schedule() to block. */
 733                         __set_current_state(TASK_UNINTERRUPTIBLE);
 734                         set_tsk_need_resched(tsk);
 735                         set_preempt_need_resched();
 736                 }
 737         }
 738 out:
 739         spin_unlock(&pfault_lock);
 740         put_task_struct(tsk);
 741 }
 742
 743 static int pfault_cpu_dead(unsigned int cpu)
 744 {
 745         struct thread_struct *thread, *next;
 746         struct task_struct *tsk;
 747
 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);
 754                 put_task_struct(tsk);
 755         }
 756         spin_unlock_irq(&pfault_lock);
 757         return 0;
 758 }
 759
 760 static int __init pfault_irq_init(void)
 761 {
 762         int rc;
 763
 764         rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
 765         if (rc)
 766                 goto out_extint;
 767         rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
 768         if (rc)
 769                 goto out_pfault;
 770         irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
 771         cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
 772                                   NULL, pfault_cpu_dead);
 773         return 0;
 774
 775 out_pfault:
 776         unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
 777 out_extint:
 778         pfault_disable = 1;
 779         return rc;
 780 }
 781 early_initcall(pfault_irq_init);
 782
 783 #endif /* CONFIG_PFAULT */
 784
 785 #if IS_ENABLED(CONFIG_PGSTE)
 786 void do_secure_storage_access(struct pt_regs *regs)
 787 {
 788         unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
 789         struct vm_area_struct *vma;
 790         struct mm_struct *mm;
 791         struct page *page;
 792         int rc;
 793
 794         switch (get_fault_type(regs)) {
 795         case USER_FAULT:
 796                 mm = current->mm;
 797                 mmap_read_lock(mm);
 798                 vma = find_vma(mm, addr);
 799                 if (!vma) {
 800                         mmap_read_unlock(mm);
 801                         do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
 802                         break;
 803                 }
 804                 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
 805                 if (IS_ERR_OR_NULL(page)) {
 806                         mmap_read_unlock(mm);
 807                         break;
 808                 }
 809                 if (arch_make_page_accessible(page))
 810                         send_sig(SIGSEGV, current, 0);
 811                 put_page(page);
 812                 mmap_read_unlock(mm);
 813                 break;
 814         case KERNEL_FAULT:
 815                 page = phys_to_page(addr);
 816                 if (unlikely(!try_get_page(page)))
 817                         break;
 818                 rc = arch_make_page_accessible(page);
 819                 put_page(page);
 820                 if (rc)
 821                         BUG();
 822                 break;
 823         case GMAP_FAULT:
 824         default:
 825                 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
 826                 WARN_ON_ONCE(1);
 827         }
 828 }
 829 NOKPROBE_SYMBOL(do_secure_storage_access);
 830
 831 void do_non_secure_storage_access(struct pt_regs *regs)
 832 {
 833         unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
 834         struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
 835
 836         if (get_fault_type(regs) != GMAP_FAULT) {
 837                 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
 838                 WARN_ON_ONCE(1);
 839                 return;
 840         }
 841
 842         if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
 843                 send_sig(SIGSEGV, current, 0);
 844 }
 845 NOKPROBE_SYMBOL(do_non_secure_storage_access);
 846
 847 void do_secure_storage_violation(struct pt_regs *regs)
 848 {
 849         /*
 850          * Either KVM messed up the secure guest mapping or the same
 851          * page is mapped into multiple secure guests.
 852          *
 853          * This exception is only triggered when a guest 2 is running
 854          * and can therefore never occur in kernel context.
 855          */
 856         printk_ratelimited(KERN_WARNING
 857                            "Secure storage violation in task: %s, pid %d\n",
 858                            current->comm, current->pid);
 859         send_sig(SIGSEGV, current, 0);
 860 }
 861
 862 #else
 863 void do_secure_storage_access(struct pt_regs *regs)
 864 {
 865         default_trap_handler(regs);
 866 }
 867
 868 void do_non_secure_storage_access(struct pt_regs *regs)
 869 {
 870         default_trap_handler(regs);
 871 }
 872
 873 void do_secure_storage_violation(struct pt_regs *regs)
 874 {
 875         default_trap_handler(regs);
 876 }
 877 #endif