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/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>
  22 #include <linux/smp.h>
  23 #include <linux/kdebug.h>
  24 #include <linux/smp_lock.h>
  25 #include <linux/init.h>
  26 #include <linux/console.h>
  27 #include <linux/module.h>
  28 #include <linux/hardirq.h>
  29 #include <linux/kprobes.h>
  30 #include <linux/uaccess.h>
  31 #include <linux/hugetlb.h>
  32 #include <asm/system.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/s390_ext.h>
  35 #include <asm/mmu_context.h>
  36 #include "../kernel/entry.h"
  37
  38 #ifndef CONFIG_64BIT
  39 #define __FAIL_ADDR_MASK 0x7ffff000
  40 #define __FIXUP_MASK 0x7fffffff
  41 #define __SUBCODE_MASK 0x0200
  42 #define __PF_RES_FIELD 0ULL
  43 #else /* CONFIG_64BIT */
  44 #define __FAIL_ADDR_MASK -4096L
  45 #define __FIXUP_MASK ~0L
  46 #define __SUBCODE_MASK 0x0600
  47 #define __PF_RES_FIELD 0x8000000000000000ULL
  48 #endif /* CONFIG_64BIT */
  49
  50 #ifdef CONFIG_SYSCTL
  51 extern int sysctl_userprocess_debug;
  52 #endif
  53
  54 #ifdef CONFIG_KPROBES
  55 static inline int notify_page_fault(struct pt_regs *regs, long err)
  56 {
  57         int ret = 0;
  58
  59         /* kprobe_running() needs smp_processor_id() */
  60         if (!user_mode(regs)) {
  61                 preempt_disable();
  62                 if (kprobe_running() && kprobe_fault_handler(regs, 14))
  63                         ret = 1;
  64                 preempt_enable();
  65         }
  66
  67         return ret;
  68 }
  69 #else
  70 static inline int notify_page_fault(struct pt_regs *regs, long err)
  71 {
  72         return 0;
  73 }
  74 #endif
  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, 1 for user space and
 103  * 2 for code execution in user space with noexec=on.
 104  */
 105 static inline int check_space(struct task_struct *tsk)
 106 {
 107         /*
 108          * The lowest two bits of S390_lowcore.trans_exc_code
 109          * indicate which paging table was used.
 110          */
 111         int desc = S390_lowcore.trans_exc_code & 3;
 112
 113         if (desc == 3)  /* Home Segment Table Descriptor */
 114                 return switch_amode == 0;
 115         if (desc == 2)  /* Secondary Segment Table Descriptor */
 116                 return tsk->thread.mm_segment.ar4;
 117 #ifdef CONFIG_S390_SWITCH_AMODE
 118         if (unlikely(desc == 1)) { /* STD determined via access register */
 119                 /* %a0 always indicates primary space. */
 120                 if (S390_lowcore.exc_access_id != 0) {
 121                         save_access_regs(tsk->thread.acrs);
 122                         /*
 123                          * An alet of 0 indicates primary space.
 124                          * An alet of 1 indicates secondary space.
 125                          * Any other alet values generate an
 126                          * alen-translation exception.
 127                          */
 128                         if (tsk->thread.acrs[S390_lowcore.exc_access_id])
 129                                 return tsk->thread.mm_segment.ar4;
 130                 }
 131         }
 132 #endif
 133         /* Primary Segment Table Descriptor */
 134         return switch_amode << s390_noexec;
 135 }
 136
 137 /*
 138  * Send SIGSEGV to task.  This is an external routine
 139  * to keep the stack usage of do_page_fault small.
 140  */
 141 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
 142                        int si_code, unsigned long address)
 143 {
 144         struct siginfo si;
 145
 146 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
 147 #if defined(CONFIG_SYSCTL)
 148         if (sysctl_userprocess_debug)
 149 #endif
 150         {
 151                 printk("User process fault: interruption code 0x%lX\n",
 152                        error_code);
 153                 printk("failing address: %lX\n", address);
 154                 show_regs(regs);
 155         }
 156 #endif
 157         si.si_signo = SIGSEGV;
 158         si.si_code = si_code;
 159         si.si_addr = (void __user *) address;
 160         force_sig_info(SIGSEGV, &si, current);
 161 }
 162
 163 static void do_no_context(struct pt_regs *regs, unsigned long error_code,
 164                           unsigned long address)
 165 {
 166         const struct exception_table_entry *fixup;
 167
 168         /* Are we prepared to handle this kernel fault?  */
 169         fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
 170         if (fixup) {
 171                 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
 172                 return;
 173         }
 174
 175         /*
 176          * Oops. The kernel tried to access some bad page. We'll have to
 177          * terminate things with extreme prejudice.
 178          */
 179         if (check_space(current) == 0)
 180                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
 181                        " at virtual kernel address %p\n", (void *)address);
 182         else
 183                 printk(KERN_ALERT "Unable to handle kernel paging request"
 184                        " at virtual user address %p\n", (void *)address);
 185
 186         die("Oops", regs, error_code);
 187         do_exit(SIGKILL);
 188 }
 189
 190 static void do_low_address(struct pt_regs *regs, unsigned long error_code)
 191 {
 192         /* Low-address protection hit in kernel mode means
 193            NULL pointer write access in kernel mode.  */
 194         if (regs->psw.mask & PSW_MASK_PSTATE) {
 195                 /* Low-address protection hit in user mode 'cannot happen'. */
 196                 die ("Low-address protection", regs, error_code);
 197                 do_exit(SIGKILL);
 198         }
 199
 200         do_no_context(regs, error_code, 0);
 201 }
 202
 203 static void do_sigbus(struct pt_regs *regs, unsigned long error_code,
 204                       unsigned long address)
 205 {
 206         struct task_struct *tsk = current;
 207         struct mm_struct *mm = tsk->mm;
 208
 209         up_read(&mm->mmap_sem);
 210         /*
 211          * Send a sigbus, regardless of whether we were in kernel
 212          * or user mode.
 213          */
 214         tsk->thread.prot_addr = address;
 215         tsk->thread.trap_no = error_code;
 216         force_sig(SIGBUS, tsk);
 217
 218         /* Kernel mode? Handle exceptions or die */
 219         if (!(regs->psw.mask & PSW_MASK_PSTATE))
 220                 do_no_context(regs, error_code, address);
 221 }
 222
 223 #ifdef CONFIG_S390_EXEC_PROTECT
 224 static int signal_return(struct mm_struct *mm, struct pt_regs *regs,
 225                          unsigned long address, unsigned long error_code)
 226 {
 227         u16 instruction;
 228         int rc;
 229 #ifdef CONFIG_COMPAT
 230         int compat;
 231 #endif
 232
 233         pagefault_disable();
 234         rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
 235         pagefault_enable();
 236         if (rc)
 237                 return -EFAULT;
 238
 239         up_read(&mm->mmap_sem);
 240         clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
 241 #ifdef CONFIG_COMPAT
 242         compat = test_tsk_thread_flag(current, TIF_31BIT);
 243         if (compat && instruction == 0x0a77)
 244                 sys32_sigreturn();
 245         else if (compat && instruction == 0x0aad)
 246                 sys32_rt_sigreturn();
 247         else
 248 #endif
 249         if (instruction == 0x0a77)
 250                 sys_sigreturn();
 251         else if (instruction == 0x0aad)
 252                 sys_rt_sigreturn();
 253         else {
 254                 current->thread.prot_addr = address;
 255                 current->thread.trap_no = error_code;
 256                 do_sigsegv(regs, error_code, SEGV_MAPERR, address);
 257         }
 258         return 0;
 259 }
 260 #endif /* CONFIG_S390_EXEC_PROTECT */
 261
 262 /*
 263  * This routine handles page faults.  It determines the address,
 264  * and the problem, and then passes it off to one of the appropriate
 265  * routines.
 266  *
 267  * error_code:
 268  *   04       Protection           ->  Write-Protection  (suprression)
 269  *   10       Segment translation  ->  Not present       (nullification)
 270  *   11       Page translation     ->  Not present       (nullification)
 271  *   3b       Region third trans.  ->  Not present       (nullification)
 272  */
 273 static inline void
 274 do_exception(struct pt_regs *regs, unsigned long error_code, int write)
 275 {
 276         struct task_struct *tsk;
 277         struct mm_struct *mm;
 278         struct vm_area_struct *vma;
 279         unsigned long address;
 280         int space;
 281         int si_code;
 282         int fault;
 283
 284         if (notify_page_fault(regs, error_code))
 285                 return;
 286
 287         tsk = current;
 288         mm = tsk->mm;
 289
 290         /* get the failing address and the affected space */
 291         address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
 292         space = check_space(tsk);
 293
 294         /*
 295          * Verify that the fault happened in user space, that
 296          * we are not in an interrupt and that there is a
 297          * user context.
 298          */
 299         if (unlikely(space == 0 || in_atomic() || !mm))
 300                 goto no_context;
 301
 302         /*
 303          * When we get here, the fault happened in the current
 304          * task's user address space, so we can switch on the
 305          * interrupts again and then search the VMAs
 306          */
 307         local_irq_enable();
 308
 309         down_read(&mm->mmap_sem);
 310
 311         si_code = SEGV_MAPERR;
 312         vma = find_vma(mm, address);
 313         if (!vma)
 314                 goto bad_area;
 315
 316 #ifdef CONFIG_S390_EXEC_PROTECT
 317         if (unlikely((space == 2) && !(vma->vm_flags & VM_EXEC)))
 318                 if (!signal_return(mm, regs, address, error_code))
 319                         /*
 320                          * signal_return() has done an up_read(&mm->mmap_sem)
 321                          * if it returns 0.
 322                          */
 323                         return;
 324 #endif
 325
 326         if (vma->vm_start <= address)
 327                 goto good_area;
 328         if (!(vma->vm_flags & VM_GROWSDOWN))
 329                 goto bad_area;
 330         if (expand_stack(vma, address))
 331                 goto bad_area;
 332 /*
 333  * Ok, we have a good vm_area for this memory access, so
 334  * we can handle it..
 335  */
 336 good_area:
 337         si_code = SEGV_ACCERR;
 338         if (!write) {
 339                 /* page not present, check vm flags */
 340                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 341                         goto bad_area;
 342         } else {
 343                 if (!(vma->vm_flags & VM_WRITE))
 344                         goto bad_area;
 345         }
 346
 347         if (is_vm_hugetlb_page(vma))
 348                 address &= HPAGE_MASK;
 349         /*
 350          * If for any reason at all we couldn't handle the fault,
 351          * make sure we exit gracefully rather than endlessly redo
 352          * the fault.
 353          */
 354         fault = handle_mm_fault(mm, vma, address, write);
 355         if (unlikely(fault & VM_FAULT_ERROR)) {
 356                 if (fault & VM_FAULT_OOM) {
 357                         up_read(&mm->mmap_sem);
 358                         pagefault_out_of_memory();
 359                         return;
 360                 } else if (fault & VM_FAULT_SIGBUS) {
 361                         do_sigbus(regs, error_code, address);
 362                         return;
 363                 }
 364                 BUG();
 365         }
 366         if (fault & VM_FAULT_MAJOR)
 367                 tsk->maj_flt++;
 368         else
 369                 tsk->min_flt++;
 370
 371         up_read(&mm->mmap_sem);
 372         /*
 373          * The instruction that caused the program check will
 374          * be repeated. Don't signal single step via SIGTRAP.
 375          */
 376         clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP);
 377         return;
 378
 379 /*
 380  * Something tried to access memory that isn't in our memory map..
 381  * Fix it, but check if it's kernel or user first..
 382  */
 383 bad_area:
 384         up_read(&mm->mmap_sem);
 385
 386         /* User mode accesses just cause a SIGSEGV */
 387         if (regs->psw.mask & PSW_MASK_PSTATE) {
 388                 tsk->thread.prot_addr = address;
 389                 tsk->thread.trap_no = error_code;
 390                 do_sigsegv(regs, error_code, si_code, address);
 391                 return;
 392         }
 393
 394 no_context:
 395         do_no_context(regs, error_code, address);
 396 }
 397
 398 void __kprobes do_protection_exception(struct pt_regs *regs,
 399                                        long error_code)
 400 {
 401         /* Protection exception is supressing, decrement psw address. */
 402         regs->psw.addr -= (error_code >> 16);
 403         /*
 404          * Check for low-address protection.  This needs to be treated
 405          * as a special case because the translation exception code
 406          * field is not guaranteed to contain valid data in this case.
 407          */
 408         if (unlikely(!(S390_lowcore.trans_exc_code & 4))) {
 409                 do_low_address(regs, error_code);
 410                 return;
 411         }
 412         do_exception(regs, 4, 1);
 413 }
 414
 415 void __kprobes do_dat_exception(struct pt_regs *regs, long error_code)
 416 {
 417         do_exception(regs, error_code & 0xff, 0);
 418 }
 419
 420 #ifdef CONFIG_64BIT
 421 void __kprobes do_asce_exception(struct pt_regs *regs, unsigned long error_code)
 422 {
 423         struct mm_struct *mm;
 424         struct vm_area_struct *vma;
 425         unsigned long address;
 426         int space;
 427
 428         mm = current->mm;
 429         address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
 430         space = check_space(current);
 431
 432         if (unlikely(space == 0 || in_atomic() || !mm))
 433                 goto no_context;
 434
 435         local_irq_enable();
 436
 437         down_read(&mm->mmap_sem);
 438         vma = find_vma(mm, address);
 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                 current->thread.prot_addr = address;
 449                 current->thread.trap_no = error_code;
 450                 do_sigsegv(regs, error_code, SEGV_MAPERR, address);
 451                 return;
 452         }
 453
 454 no_context:
 455         do_no_context(regs, error_code, address);
 456 }
 457 #endif
 458
 459 #ifdef CONFIG_PFAULT
 460 /*
 461  * 'pfault' pseudo page faults routines.
 462  */
 463 static ext_int_info_t ext_int_pfault;
 464 static int pfault_disable = 0;
 465
 466 static int __init nopfault(char *str)
 467 {
 468         pfault_disable = 1;
 469         return 1;
 470 }
 471
 472 __setup("nopfault", nopfault);
 473
 474 typedef struct {
 475         __u16 refdiagc;
 476         __u16 reffcode;
 477         __u16 refdwlen;
 478         __u16 refversn;
 479         __u64 refgaddr;
 480         __u64 refselmk;
 481         __u64 refcmpmk;
 482         __u64 reserved;
 483 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
 484
 485 int pfault_init(void)
 486 {
 487         pfault_refbk_t refbk =
 488                 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
 489                   __PF_RES_FIELD };
 490         int rc;
 491
 492         if (!MACHINE_IS_VM || pfault_disable)
 493                 return -1;
 494         asm volatile(
 495                 "       diag    %1,%0,0x258\n"
 496                 "0:     j       2f\n"
 497                 "1:     la      %0,8\n"
 498                 "2:\n"
 499                 EX_TABLE(0b,1b)
 500                 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
 501         __ctl_set_bit(0, 9);
 502         return rc;
 503 }
 504
 505 void pfault_fini(void)
 506 {
 507         pfault_refbk_t refbk =
 508         { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
 509
 510         if (!MACHINE_IS_VM || pfault_disable)
 511                 return;
 512         __ctl_clear_bit(0,9);
 513         asm volatile(
 514                 "       diag    %0,0,0x258\n"
 515                 "0:\n"
 516                 EX_TABLE(0b,0b)
 517                 : : "a" (&refbk), "m" (refbk) : "cc");
 518 }
 519
 520 static void pfault_interrupt(__u16 error_code)
 521 {
 522         struct task_struct *tsk;
 523         __u16 subcode;
 524
 525         /*
 526          * Get the external interruption subcode & pfault
 527          * initial/completion signal bit. VM stores this
 528          * in the 'cpu address' field associated with the
 529          * external interrupt.
 530          */
 531         subcode = S390_lowcore.cpu_addr;
 532         if ((subcode & 0xff00) != __SUBCODE_MASK)
 533                 return;
 534
 535         /*
 536          * Get the token (= address of the task structure of the affected task).
 537          */
 538         tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
 539
 540         if (subcode & 0x0080) {
 541                 /* signal bit is set -> a page has been swapped in by VM */
 542                 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
 543                         /* Initial interrupt was faster than the completion
 544                          * interrupt. pfault_wait is valid. Set pfault_wait
 545                          * back to zero and wake up the process. This can
 546                          * safely be done because the task is still sleeping
 547                          * and can't produce new pfaults. */
 548                         tsk->thread.pfault_wait = 0;
 549                         wake_up_process(tsk);
 550                         put_task_struct(tsk);
 551                 }
 552         } else {
 553                 /* signal bit not set -> a real page is missing. */
 554                 get_task_struct(tsk);
 555                 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 556                 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
 557                         /* Completion interrupt was faster than the initial
 558                          * interrupt (swapped in a -1 for pfault_wait). Set
 559                          * pfault_wait back to zero and exit. This can be
 560                          * done safely because tsk is running in kernel
 561                          * mode and can't produce new pfaults. */
 562                         tsk->thread.pfault_wait = 0;
 563                         set_task_state(tsk, TASK_RUNNING);
 564                         put_task_struct(tsk);
 565                 } else
 566                         set_tsk_need_resched(tsk);
 567         }
 568 }
 569
 570 void __init pfault_irq_init(void)
 571 {
 572         if (!MACHINE_IS_VM)
 573                 return;
 574
 575         /*
 576          * Try to get pfault pseudo page faults going.
 577          */
 578         if (register_early_external_interrupt(0x2603, pfault_interrupt,
 579                                               &ext_int_pfault) != 0)
 580                 panic("Couldn't request external interrupt 0x2603");
 581
 582         if (pfault_init() == 0)
 583                 return;
 584
 585         /* Tough luck, no pfault. */
 586         pfault_disable = 1;
 587         unregister_early_external_interrupt(0x2603, pfault_interrupt,
 588                                             &ext_int_pfault);
 589 }
 590 #endif