arch/s390/mm/fault.c

   1 /*
   2  *  arch/s390/mm/fault.c
   3  *
   4  *  S390 version
   5  *    Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
   6  *    Author(s): Hartmut Penner (hp@de.ibm.com)
   7  *               Ulrich Weigand (uweigand@de.ibm.com)
   8  *
   9  *  Derived from "arch/i386/mm/fault.c"
  10  *    Copyright (C) 1995  Linus Torvalds
  11  */
  12
  13 #include <linux/kernel_stat.h>
  14 #include <linux/perf_event.h>
  15 #include <linux/signal.h>
  16 #include <linux/sched.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/module.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/system.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/s390_ext.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/compat.h>
  40 #include "../kernel/entry.h"
  41
  42 #ifndef CONFIG_64BIT
  43 #define __FAIL_ADDR_MASK 0x7ffff000
  44 #define __SUBCODE_MASK 0x0200
  45 #define __PF_RES_FIELD 0ULL
  46 #else /* CONFIG_64BIT */
  47 #define __FAIL_ADDR_MASK -4096L
  48 #define __SUBCODE_MASK 0x0600
  49 #define __PF_RES_FIELD 0x8000000000000000ULL
  50 #endif /* CONFIG_64BIT */
  51
  52 #define VM_FAULT_BADCONTEXT     0x010000
  53 #define VM_FAULT_BADMAP         0x020000
  54 #define VM_FAULT_BADACCESS      0x040000
  55
  56 static unsigned long store_indication;
  57
  58 void fault_init(void)
  59 {
  60         if (test_facility(2) && test_facility(75))
  61                 store_indication = 0xc00;
  62 }
  63
  64 static inline int notify_page_fault(struct pt_regs *regs)
  65 {
  66         int ret = 0;
  67
  68         /* kprobe_running() needs smp_processor_id() */
  69         if (kprobes_built_in() && !user_mode(regs)) {
  70                 preempt_disable();
  71                 if (kprobe_running() && kprobe_fault_handler(regs, 14))
  72                         ret = 1;
  73                 preempt_enable();
  74         }
  75         return ret;
  76 }
  77
  78
  79 /*
  80  * Unlock any spinlocks which will prevent us from getting the
  81  * message out.
  82  */
  83 void bust_spinlocks(int yes)
  84 {
  85         if (yes) {
  86                 oops_in_progress = 1;
  87         } else {
  88                 int loglevel_save = console_loglevel;
  89                 console_unblank();
  90                 oops_in_progress = 0;
  91                 /*
  92                  * OK, the message is on the console.  Now we call printk()
  93                  * without oops_in_progress set so that printk will give klogd
  94                  * a poke.  Hold onto your hats...
  95                  */
  96                 console_loglevel = 15;
  97                 printk(" ");
  98                 console_loglevel = loglevel_save;
  99         }
 100 }
 101
 102 /*
 103  * Returns the address space associated with the fault.
 104  * Returns 0 for kernel space and 1 for user space.
 105  */
 106 static inline int user_space_fault(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 &= 3;
 113         if (trans_exc_code == 2)
 114                 /* Access via secondary space, set_fs setting decides */
 115                 return current->thread.mm_segment.ar4;
 116         if (user_mode == HOME_SPACE_MODE)
 117                 /* User space if the access has been done via home space. */
 118                 return trans_exc_code == 3;
 119         /*
 120          * If the user space is not the home space the kernel runs in home
 121          * space. Access via secondary space has already been covered,
 122          * access via primary space or access register is from user space
 123          * and access via home space is from the kernel.
 124          */
 125         return trans_exc_code != 3;
 126 }
 127
 128 static inline void report_user_fault(struct pt_regs *regs, long int_code,
 129                                      int signr, unsigned long address)
 130 {
 131         if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
 132                 return;
 133         if (!unhandled_signal(current, signr))
 134                 return;
 135         if (!printk_ratelimit())
 136                 return;
 137         printk("User process fault: interruption code 0x%lX ", int_code);
 138         print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
 139         printk("\n");
 140         printk("failing address: %lX\n", address);
 141         show_regs(regs);
 142 }
 143
 144 /*
 145  * Send SIGSEGV to task.  This is an external routine
 146  * to keep the stack usage of do_page_fault small.
 147  */
 148 static noinline void do_sigsegv(struct pt_regs *regs, long int_code,
 149                                 int si_code, unsigned long trans_exc_code)
 150 {
 151         struct siginfo si;
 152         unsigned long address;
 153
 154         address = trans_exc_code & __FAIL_ADDR_MASK;
 155         current->thread.prot_addr = address;
 156         current->thread.trap_no = int_code;
 157         report_user_fault(regs, int_code, SIGSEGV, address);
 158         si.si_signo = SIGSEGV;
 159         si.si_code = si_code;
 160         si.si_addr = (void __user *) address;
 161         force_sig_info(SIGSEGV, &si, current);
 162 }
 163
 164 static noinline void do_no_context(struct pt_regs *regs, long int_code,
 165                                    unsigned long trans_exc_code)
 166 {
 167         const struct exception_table_entry *fixup;
 168         unsigned long address;
 169
 170         /* Are we prepared to handle this kernel fault?  */
 171         fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
 172         if (fixup) {
 173                 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
 174                 return;
 175         }
 176
 177         /*
 178          * Oops. The kernel tried to access some bad page. We'll have to
 179          * terminate things with extreme prejudice.
 180          */
 181         address = trans_exc_code & __FAIL_ADDR_MASK;
 182         if (!user_space_fault(trans_exc_code))
 183                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
 184                        " at virtual kernel address %p\n", (void *)address);
 185         else
 186                 printk(KERN_ALERT "Unable to handle kernel paging request"
 187                        " at virtual user address %p\n", (void *)address);
 188
 189         die("Oops", regs, int_code);
 190         do_exit(SIGKILL);
 191 }
 192
 193 static noinline void do_low_address(struct pt_regs *regs, long int_code,
 194                                     unsigned long trans_exc_code)
 195 {
 196         /* Low-address protection hit in kernel mode means
 197            NULL pointer write access in kernel mode.  */
 198         if (regs->psw.mask & PSW_MASK_PSTATE) {
 199                 /* Low-address protection hit in user mode 'cannot happen'. */
 200                 die ("Low-address protection", regs, int_code);
 201                 do_exit(SIGKILL);
 202         }
 203
 204         do_no_context(regs, int_code, trans_exc_code);
 205 }
 206
 207 static noinline void do_sigbus(struct pt_regs *regs, long int_code,
 208                                unsigned long trans_exc_code)
 209 {
 210         struct task_struct *tsk = current;
 211         unsigned long address;
 212         struct siginfo si;
 213
 214         /*
 215          * Send a sigbus, regardless of whether we were in kernel
 216          * or user mode.
 217          */
 218         address = trans_exc_code & __FAIL_ADDR_MASK;
 219         tsk->thread.prot_addr = address;
 220         tsk->thread.trap_no = int_code;
 221         si.si_signo = SIGBUS;
 222         si.si_errno = 0;
 223         si.si_code = BUS_ADRERR;
 224         si.si_addr = (void __user *) address;
 225         force_sig_info(SIGBUS, &si, tsk);
 226 }
 227
 228 #ifdef CONFIG_S390_EXEC_PROTECT
 229 static noinline int signal_return(struct pt_regs *regs, long int_code,
 230                                   unsigned long trans_exc_code)
 231 {
 232         u16 instruction;
 233         int rc;
 234
 235         rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
 236
 237         if (!rc && instruction == 0x0a77) {
 238                 clear_tsk_thread_flag(current, TIF_PER_TRAP);
 239                 if (is_compat_task())
 240                         sys32_sigreturn();
 241                 else
 242                         sys_sigreturn();
 243         } else if (!rc && instruction == 0x0aad) {
 244                 clear_tsk_thread_flag(current, TIF_PER_TRAP);
 245                 if (is_compat_task())
 246                         sys32_rt_sigreturn();
 247                 else
 248                         sys_rt_sigreturn();
 249         } else
 250                 do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code);
 251         return 0;
 252 }
 253 #endif /* CONFIG_S390_EXEC_PROTECT */
 254
 255 static noinline void do_fault_error(struct pt_regs *regs, long int_code,
 256                                     unsigned long trans_exc_code, int fault)
 257 {
 258         int si_code;
 259
 260         switch (fault) {
 261         case VM_FAULT_BADACCESS:
 262 #ifdef CONFIG_S390_EXEC_PROTECT
 263                 if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
 264                     (trans_exc_code & 3) == 0) {
 265                         signal_return(regs, int_code, trans_exc_code);
 266                         break;
 267                 }
 268 #endif /* CONFIG_S390_EXEC_PROTECT */
 269         case VM_FAULT_BADMAP:
 270                 /* Bad memory access. Check if it is kernel or user space. */
 271                 if (regs->psw.mask & PSW_MASK_PSTATE) {
 272                         /* User mode accesses just cause a SIGSEGV */
 273                         si_code = (fault == VM_FAULT_BADMAP) ?
 274                                 SEGV_MAPERR : SEGV_ACCERR;
 275                         do_sigsegv(regs, int_code, si_code, trans_exc_code);
 276                         return;
 277                 }
 278         case VM_FAULT_BADCONTEXT:
 279                 do_no_context(regs, int_code, trans_exc_code);
 280                 break;
 281         default: /* fault & VM_FAULT_ERROR */
 282                 if (fault & VM_FAULT_OOM)
 283                         pagefault_out_of_memory();
 284                 else if (fault & VM_FAULT_SIGBUS) {
 285                         /* Kernel mode? Handle exceptions or die */
 286                         if (!(regs->psw.mask & PSW_MASK_PSTATE))
 287                                 do_no_context(regs, int_code, trans_exc_code);
 288                         else
 289                                 do_sigbus(regs, int_code, trans_exc_code);
 290                 } else
 291                         BUG();
 292                 break;
 293         }
 294 }
 295
 296 /*
 297  * This routine handles page faults.  It determines the address,
 298  * and the problem, and then passes it off to one of the appropriate
 299  * routines.
 300  *
 301  * interruption code (int_code):
 302  *   04       Protection           ->  Write-Protection  (suprression)
 303  *   10       Segment translation  ->  Not present       (nullification)
 304  *   11       Page translation     ->  Not present       (nullification)
 305  *   3b       Region third trans.  ->  Not present       (nullification)
 306  */
 307 static inline int do_exception(struct pt_regs *regs, int access,
 308                                unsigned long trans_exc_code)
 309 {
 310         struct task_struct *tsk;
 311         struct mm_struct *mm;
 312         struct vm_area_struct *vma;
 313         unsigned long address;
 314         int fault, write;
 315
 316         if (notify_page_fault(regs))
 317                 return 0;
 318
 319         tsk = current;
 320         mm = tsk->mm;
 321
 322         /*
 323          * Verify that the fault happened in user space, that
 324          * we are not in an interrupt and that there is a
 325          * user context.
 326          */
 327         fault = VM_FAULT_BADCONTEXT;
 328         if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
 329                 goto out;
 330
 331         address = trans_exc_code & __FAIL_ADDR_MASK;
 332         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
 333         down_read(&mm->mmap_sem);
 334
 335         fault = VM_FAULT_BADMAP;
 336         vma = find_vma(mm, address);
 337         if (!vma)
 338                 goto out_up;
 339
 340         if (unlikely(vma->vm_start > address)) {
 341                 if (!(vma->vm_flags & VM_GROWSDOWN))
 342                         goto out_up;
 343                 if (expand_stack(vma, address))
 344                         goto out_up;
 345         }
 346
 347         /*
 348          * Ok, we have a good vm_area for this memory access, so
 349          * we can handle it..
 350          */
 351         fault = VM_FAULT_BADACCESS;
 352         if (unlikely(!(vma->vm_flags & access)))
 353                 goto out_up;
 354
 355         if (is_vm_hugetlb_page(vma))
 356                 address &= HPAGE_MASK;
 357         /*
 358          * If for any reason at all we couldn't handle the fault,
 359          * make sure we exit gracefully rather than endlessly redo
 360          * the fault.
 361          */
 362         write = (access == VM_WRITE ||
 363                  (trans_exc_code & store_indication) == 0x400) ?
 364                 FAULT_FLAG_WRITE : 0;
 365         fault = handle_mm_fault(mm, vma, address, write);
 366         if (unlikely(fault & VM_FAULT_ERROR))
 367                 goto out_up;
 368
 369         if (fault & VM_FAULT_MAJOR) {
 370                 tsk->maj_flt++;
 371                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
 372                                      regs, address);
 373         } else {
 374                 tsk->min_flt++;
 375                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
 376                                      regs, address);
 377         }
 378         /*
 379          * The instruction that caused the program check will
 380          * be repeated. Don't signal single step via SIGTRAP.
 381          */
 382         clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
 383         fault = 0;
 384 out_up:
 385         up_read(&mm->mmap_sem);
 386 out:
 387         return fault;
 388 }
 389
 390 void __kprobes do_protection_exception(struct pt_regs *regs, long pgm_int_code,
 391                                        unsigned long trans_exc_code)
 392 {
 393         int fault;
 394
 395         /* Protection exception is suppressing, decrement psw address. */
 396         regs->psw.addr -= (pgm_int_code >> 16);
 397         /*
 398          * Check for low-address protection.  This needs to be treated
 399          * as a special case because the translation exception code
 400          * field is not guaranteed to contain valid data in this case.
 401          */
 402         if (unlikely(!(trans_exc_code & 4))) {
 403                 do_low_address(regs, pgm_int_code, trans_exc_code);
 404                 return;
 405         }
 406         fault = do_exception(regs, VM_WRITE, trans_exc_code);
 407         if (unlikely(fault))
 408                 do_fault_error(regs, 4, trans_exc_code, fault);
 409 }
 410
 411 void __kprobes do_dat_exception(struct pt_regs *regs, long pgm_int_code,
 412                                 unsigned long trans_exc_code)
 413 {
 414         int access, fault;
 415
 416         access = VM_READ | VM_EXEC | VM_WRITE;
 417 #ifdef CONFIG_S390_EXEC_PROTECT
 418         if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
 419             (trans_exc_code & 3) == 0)
 420                 access = VM_EXEC;
 421 #endif
 422         fault = do_exception(regs, access, trans_exc_code);
 423         if (unlikely(fault))
 424                 do_fault_error(regs, pgm_int_code & 255, trans_exc_code, fault);
 425 }
 426
 427 #ifdef CONFIG_64BIT
 428 void __kprobes do_asce_exception(struct pt_regs *regs, long pgm_int_code,
 429                                  unsigned long trans_exc_code)
 430 {
 431         struct mm_struct *mm = current->mm;
 432         struct vm_area_struct *vma;
 433
 434         if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
 435                 goto no_context;
 436
 437         down_read(&mm->mmap_sem);
 438         vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
 439         up_read(&mm->mmap_sem);
 440
 441         if (vma) {
 442                 update_mm(mm, current);
 443                 return;
 444         }
 445
 446         /* User mode accesses just cause a SIGSEGV */
 447         if (regs->psw.mask & PSW_MASK_PSTATE) {
 448                 do_sigsegv(regs, pgm_int_code, SEGV_MAPERR, trans_exc_code);
 449                 return;
 450         }
 451
 452 no_context:
 453         do_no_context(regs, pgm_int_code, trans_exc_code);
 454 }
 455 #endif
 456
 457 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
 458 {
 459         struct pt_regs regs;
 460         int access, fault;
 461
 462         regs.psw.mask = psw_kernel_bits;
 463         if (!irqs_disabled())
 464                 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
 465         regs.psw.addr = (unsigned long) __builtin_return_address(0);
 466         regs.psw.addr |= PSW_ADDR_AMODE;
 467         uaddr &= PAGE_MASK;
 468         access = write ? VM_WRITE : VM_READ;
 469         fault = do_exception(&regs, access, uaddr | 2);
 470         if (unlikely(fault)) {
 471                 if (fault & VM_FAULT_OOM) {
 472                         pagefault_out_of_memory();
 473                         fault = 0;
 474                 } else if (fault & VM_FAULT_SIGBUS)
 475                         do_sigbus(&regs, pgm_int_code, uaddr);
 476         }
 477         return fault ? -EFAULT : 0;
 478 }
 479
 480 #ifdef CONFIG_PFAULT
 481 /*
 482  * 'pfault' pseudo page faults routines.
 483  */
 484 static int pfault_disable;
 485
 486 static int __init nopfault(char *str)
 487 {
 488         pfault_disable = 1;
 489         return 1;
 490 }
 491
 492 __setup("nopfault", nopfault);
 493
 494 typedef struct {
 495         __u16 refdiagc;
 496         __u16 reffcode;
 497         __u16 refdwlen;
 498         __u16 refversn;
 499         __u64 refgaddr;
 500         __u64 refselmk;
 501         __u64 refcmpmk;
 502         __u64 reserved;
 503 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
 504
 505 int pfault_init(void)
 506 {
 507         pfault_refbk_t refbk =
 508                 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
 509                   __PF_RES_FIELD };
 510         int rc;
 511
 512         if (!MACHINE_IS_VM || pfault_disable)
 513                 return -1;
 514         asm volatile(
 515                 "       diag    %1,%0,0x258\n"
 516                 "0:     j       2f\n"
 517                 "1:     la      %0,8\n"
 518                 "2:\n"
 519                 EX_TABLE(0b,1b)
 520                 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
 521         __ctl_set_bit(0, 9);
 522         return rc;
 523 }
 524
 525 void pfault_fini(void)
 526 {
 527         pfault_refbk_t refbk =
 528         { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
 529
 530         if (!MACHINE_IS_VM || pfault_disable)
 531                 return;
 532         __ctl_clear_bit(0,9);
 533         asm volatile(
 534                 "       diag    %0,0,0x258\n"
 535                 "0:\n"
 536                 EX_TABLE(0b,0b)
 537                 : : "a" (&refbk), "m" (refbk) : "cc");
 538 }
 539
 540 static void pfault_interrupt(unsigned int ext_int_code,
 541                              unsigned int param32, unsigned long param64)
 542 {
 543         struct task_struct *tsk;
 544         __u16 subcode;
 545
 546         kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
 547         /*
 548          * Get the external interruption subcode & pfault
 549          * initial/completion signal bit. VM stores this
 550          * in the 'cpu address' field associated with the
 551          * external interrupt.
 552          */
 553         subcode = ext_int_code >> 16;
 554         if ((subcode & 0xff00) != __SUBCODE_MASK)
 555                 return;
 556
 557         /*
 558          * Get the token (= address of the task structure of the affected task).
 559          */
 560 #ifdef CONFIG_64BIT
 561         tsk = *(struct task_struct **) param64;
 562 #else
 563         tsk = *(struct task_struct **) param32;
 564 #endif
 565
 566         if (subcode & 0x0080) {
 567                 /* signal bit is set -> a page has been swapped in by VM */
 568                 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
 569                         /* Initial interrupt was faster than the completion
 570                          * interrupt. pfault_wait is valid. Set pfault_wait
 571                          * back to zero and wake up the process. This can
 572                          * safely be done because the task is still sleeping
 573                          * and can't produce new pfaults. */
 574                         tsk->thread.pfault_wait = 0;
 575                         wake_up_process(tsk);
 576                         put_task_struct(tsk);
 577                 }
 578         } else {
 579                 /* signal bit not set -> a real page is missing. */
 580                 get_task_struct(tsk);
 581                 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 582                 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
 583                         /* Completion interrupt was faster than the initial
 584                          * interrupt (swapped in a -1 for pfault_wait). Set
 585                          * pfault_wait back to zero and exit. This can be
 586                          * done safely because tsk is running in kernel
 587                          * mode and can't produce new pfaults. */
 588                         tsk->thread.pfault_wait = 0;
 589                         set_task_state(tsk, TASK_RUNNING);
 590                         put_task_struct(tsk);
 591                 } else
 592                         set_tsk_need_resched(tsk);
 593         }
 594 }
 595
 596 static int __init pfault_irq_init(void)
 597 {
 598         int rc;
 599
 600         if (!MACHINE_IS_VM)
 601                 return 0;
 602         /*
 603          * Try to get pfault pseudo page faults going.
 604          */
 605         rc = register_external_interrupt(0x2603, pfault_interrupt);
 606         if (rc) {
 607                 pfault_disable = 1;
 608                 return rc;
 609         }
 610         if (pfault_init() == 0)
 611                 return 0;
 612
 613         /* Tough luck, no pfault. */
 614         pfault_disable = 1;
 615         unregister_external_interrupt(0x2603, pfault_interrupt);
 616         return 0;
 617 }
 618 early_initcall(pfault_irq_init);
 619
 620 #endif