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/smp_lock.h>
  24 #include <linux/init.h>
  25 #include <linux/console.h>
  26 #include <linux/module.h>
  27 #include <linux/hardirq.h>
  28 #include <linux/kprobes.h>
  29
  30 #include <asm/system.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/pgtable.h>
  33 #include <asm/kdebug.h>
  34 #include <asm/s390_ext.h>
  35
  36 #ifndef CONFIG_64BIT
  37 #define __FAIL_ADDR_MASK 0x7ffff000
  38 #define __FIXUP_MASK 0x7fffffff
  39 #define __SUBCODE_MASK 0x0200
  40 #define __PF_RES_FIELD 0ULL
  41 #else /* CONFIG_64BIT */
  42 #define __FAIL_ADDR_MASK -4096L
  43 #define __FIXUP_MASK ~0L
  44 #define __SUBCODE_MASK 0x0600
  45 #define __PF_RES_FIELD 0x8000000000000000ULL
  46 #endif /* CONFIG_64BIT */
  47
  48 #ifdef CONFIG_SYSCTL
  49 extern int sysctl_userprocess_debug;
  50 #endif
  51
  52 extern void die(const char *,struct pt_regs *,long);
  53
  54 #ifdef CONFIG_KPROBES
  55 static ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
  56 int register_page_fault_notifier(struct notifier_block *nb)
  57 {
  58         return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
  59 }
  60
  61 int unregister_page_fault_notifier(struct notifier_block *nb)
  62 {
  63         return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
  64 }
  65
  66 static inline int notify_page_fault(enum die_val val, const char *str,
  67                         struct pt_regs *regs, long err, int trap, int sig)
  68 {
  69         struct die_args args = {
  70                 .regs = regs,
  71                 .str = str,
  72                 .err = err,
  73                 .trapnr = trap,
  74                 .signr = sig
  75         };
  76         return atomic_notifier_call_chain(&notify_page_fault_chain, val, &args);
  77 }
  78 #else
  79 static inline int notify_page_fault(enum die_val val, const char *str,
  80                         struct pt_regs *regs, long err, int trap, int sig)
  81 {
  82         return NOTIFY_DONE;
  83 }
  84 #endif
  85
  86 extern spinlock_t timerlist_lock;
  87
  88 /*
  89  * Unlock any spinlocks which will prevent us from getting the
  90  * message out (timerlist_lock is acquired through the
  91  * console unblank code)
  92  */
  93 void bust_spinlocks(int yes)
  94 {
  95         if (yes) {
  96                 oops_in_progress = 1;
  97         } else {
  98                 int loglevel_save = console_loglevel;
  99                 console_unblank();
 100                 oops_in_progress = 0;
 101                 /*
 102                  * OK, the message is on the console.  Now we call printk()
 103                  * without oops_in_progress set so that printk will give klogd
 104                  * a poke.  Hold onto your hats...
 105                  */
 106                 console_loglevel = 15;
 107                 printk(" ");
 108                 console_loglevel = loglevel_save;
 109         }
 110 }
 111
 112 /*
 113  * Check which address space is addressed by the access
 114  * register in S390_lowcore.exc_access_id.
 115  * Returns 1 for user space and 0 for kernel space.
 116  */
 117 static int __check_access_register(struct pt_regs *regs, int error_code)
 118 {
 119         int areg = S390_lowcore.exc_access_id;
 120
 121         if (areg == 0)
 122                 /* Access via access register 0 -> kernel address */
 123                 return 0;
 124         save_access_regs(current->thread.acrs);
 125         if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
 126                 /*
 127                  * access register contains 0 -> kernel address,
 128                  * access register contains 1 -> user space address
 129                  */
 130                 return current->thread.acrs[areg];
 131
 132         /* Something unhealthy was done with the access registers... */
 133         die("page fault via unknown access register", regs, error_code);
 134         do_exit(SIGKILL);
 135         return 0;
 136 }
 137
 138 /*
 139  * Check which address space the address belongs to.
 140  * Returns 1 for user space and 0 for kernel space.
 141  */
 142 static inline int check_user_space(struct pt_regs *regs, int error_code)
 143 {
 144         /*
 145          * The lowest two bits of S390_lowcore.trans_exc_code indicate
 146          * which paging table was used:
 147          *   0: Primary Segment Table Descriptor
 148          *   1: STD determined via access register
 149          *   2: Secondary Segment Table Descriptor
 150          *   3: Home Segment Table Descriptor
 151          */
 152         int descriptor = S390_lowcore.trans_exc_code & 3;
 153         if (unlikely(descriptor == 1))
 154                 return __check_access_register(regs, error_code);
 155         if (descriptor == 2)
 156                 return current->thread.mm_segment.ar4;
 157         return descriptor != 0;
 158 }
 159
 160 /*
 161  * Send SIGSEGV to task.  This is an external routine
 162  * to keep the stack usage of do_page_fault small.
 163  */
 164 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
 165                        int si_code, unsigned long address)
 166 {
 167         struct siginfo si;
 168
 169 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
 170 #if defined(CONFIG_SYSCTL)
 171         if (sysctl_userprocess_debug)
 172 #endif
 173         {
 174                 printk("User process fault: interruption code 0x%lX\n",
 175                        error_code);
 176                 printk("failing address: %lX\n", address);
 177                 show_regs(regs);
 178         }
 179 #endif
 180         si.si_signo = SIGSEGV;
 181         si.si_code = si_code;
 182         si.si_addr = (void __user *) address;
 183         force_sig_info(SIGSEGV, &si, current);
 184 }
 185
 186 /*
 187  * This routine handles page faults.  It determines the address,
 188  * and the problem, and then passes it off to one of the appropriate
 189  * routines.
 190  *
 191  * error_code:
 192  *   04       Protection           ->  Write-Protection  (suprression)
 193  *   10       Segment translation  ->  Not present       (nullification)
 194  *   11       Page translation     ->  Not present       (nullification)
 195  *   3b       Region third trans.  ->  Not present       (nullification)
 196  */
 197 static inline void __kprobes
 198 do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection)
 199 {
 200         struct task_struct *tsk;
 201         struct mm_struct *mm;
 202         struct vm_area_struct * vma;
 203         unsigned long address;
 204         int user_address;
 205         const struct exception_table_entry *fixup;
 206         int si_code = SEGV_MAPERR;
 207
 208         tsk = current;
 209         mm = tsk->mm;
 210
 211         if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14,
 212                                         SIGSEGV) == NOTIFY_STOP)
 213                 return;
 214
 215         /*
 216          * Check for low-address protection.  This needs to be treated
 217          * as a special case because the translation exception code
 218          * field is not guaranteed to contain valid data in this case.
 219          */
 220         if (is_protection && !(S390_lowcore.trans_exc_code & 4)) {
 221
 222                 /* Low-address protection hit in kernel mode means
 223                    NULL pointer write access in kernel mode.  */
 224                 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
 225                         address = 0;
 226                         user_address = 0;
 227                         goto no_context;
 228                 }
 229
 230                 /* Low-address protection hit in user mode 'cannot happen'.  */
 231                 die ("Low-address protection", regs, error_code);
 232                 do_exit(SIGKILL);
 233         }
 234
 235         /*
 236          * get the failing address
 237          * more specific the segment and page table portion of
 238          * the address
 239          */
 240         address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
 241         user_address = check_user_space(regs, error_code);
 242
 243         /*
 244          * Verify that the fault happened in user space, that
 245          * we are not in an interrupt and that there is a
 246          * user context.
 247          */
 248         if (user_address == 0 || in_atomic() || !mm)
 249                 goto no_context;
 250
 251         /*
 252          * When we get here, the fault happened in the current
 253          * task's user address space, so we can switch on the
 254          * interrupts again and then search the VMAs
 255          */
 256         local_irq_enable();
 257
 258         down_read(&mm->mmap_sem);
 259
 260         vma = find_vma(mm, address);
 261         if (!vma)
 262                 goto bad_area;
 263         if (vma->vm_start <= address)
 264                 goto good_area;
 265         if (!(vma->vm_flags & VM_GROWSDOWN))
 266                 goto bad_area;
 267         if (expand_stack(vma, address))
 268                 goto bad_area;
 269 /*
 270  * Ok, we have a good vm_area for this memory access, so
 271  * we can handle it..
 272  */
 273 good_area:
 274         si_code = SEGV_ACCERR;
 275         if (!is_protection) {
 276                 /* page not present, check vm flags */
 277                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 278                         goto bad_area;
 279         } else {
 280                 if (!(vma->vm_flags & VM_WRITE))
 281                         goto bad_area;
 282         }
 283
 284 survive:
 285         /*
 286          * If for any reason at all we couldn't handle the fault,
 287          * make sure we exit gracefully rather than endlessly redo
 288          * the fault.
 289          */
 290         switch (handle_mm_fault(mm, vma, address, is_protection)) {
 291         case VM_FAULT_MINOR:
 292                 tsk->min_flt++;
 293                 break;
 294         case VM_FAULT_MAJOR:
 295                 tsk->maj_flt++;
 296                 break;
 297         case VM_FAULT_SIGBUS:
 298                 goto do_sigbus;
 299         case VM_FAULT_OOM:
 300                 goto out_of_memory;
 301         default:
 302                 BUG();
 303         }
 304
 305         up_read(&mm->mmap_sem);
 306         /*
 307          * The instruction that caused the program check will
 308          * be repeated. Don't signal single step via SIGTRAP.
 309          */
 310         clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
 311         return;
 312
 313 /*
 314  * Something tried to access memory that isn't in our memory map..
 315  * Fix it, but check if it's kernel or user first..
 316  */
 317 bad_area:
 318         up_read(&mm->mmap_sem);
 319
 320         /* User mode accesses just cause a SIGSEGV */
 321         if (regs->psw.mask & PSW_MASK_PSTATE) {
 322                 tsk->thread.prot_addr = address;
 323                 tsk->thread.trap_no = error_code;
 324                 do_sigsegv(regs, error_code, si_code, address);
 325                 return;
 326         }
 327
 328 no_context:
 329         /* Are we prepared to handle this kernel fault?  */
 330         fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
 331         if (fixup) {
 332                 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
 333                 return;
 334         }
 335
 336 /*
 337  * Oops. The kernel tried to access some bad page. We'll have to
 338  * terminate things with extreme prejudice.
 339  */
 340         if (user_address == 0)
 341                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
 342                        " at virtual kernel address %p\n", (void *)address);
 343         else
 344                 printk(KERN_ALERT "Unable to handle kernel paging request"
 345                        " at virtual user address %p\n", (void *)address);
 346
 347         die("Oops", regs, error_code);
 348         do_exit(SIGKILL);
 349
 350
 351 /*
 352  * We ran out of memory, or some other thing happened to us that made
 353  * us unable to handle the page fault gracefully.
 354 */
 355 out_of_memory:
 356         up_read(&mm->mmap_sem);
 357         if (is_init(tsk)) {
 358                 yield();
 359                 down_read(&mm->mmap_sem);
 360                 goto survive;
 361         }
 362         printk("VM: killing process %s\n", tsk->comm);
 363         if (regs->psw.mask & PSW_MASK_PSTATE)
 364                 do_exit(SIGKILL);
 365         goto no_context;
 366
 367 do_sigbus:
 368         up_read(&mm->mmap_sem);
 369
 370         /*
 371          * Send a sigbus, regardless of whether we were in kernel
 372          * or user mode.
 373          */
 374         tsk->thread.prot_addr = address;
 375         tsk->thread.trap_no = error_code;
 376         force_sig(SIGBUS, tsk);
 377
 378         /* Kernel mode? Handle exceptions or die */
 379         if (!(regs->psw.mask & PSW_MASK_PSTATE))
 380                 goto no_context;
 381 }
 382
 383 void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
 384 {
 385         regs->psw.addr -= (error_code >> 16);
 386         do_exception(regs, 4, 1);
 387 }
 388
 389 void do_dat_exception(struct pt_regs *regs, unsigned long error_code)
 390 {
 391         do_exception(regs, error_code & 0xff, 0);
 392 }
 393
 394 #ifdef CONFIG_PFAULT
 395 /*
 396  * 'pfault' pseudo page faults routines.
 397  */
 398 static ext_int_info_t ext_int_pfault;
 399 static int pfault_disable = 0;
 400
 401 static int __init nopfault(char *str)
 402 {
 403         pfault_disable = 1;
 404         return 1;
 405 }
 406
 407 __setup("nopfault", nopfault);
 408
 409 typedef struct {
 410         __u16 refdiagc;
 411         __u16 reffcode;
 412         __u16 refdwlen;
 413         __u16 refversn;
 414         __u64 refgaddr;
 415         __u64 refselmk;
 416         __u64 refcmpmk;
 417         __u64 reserved;
 418 } __attribute__ ((packed)) pfault_refbk_t;
 419
 420 int pfault_init(void)
 421 {
 422         pfault_refbk_t refbk =
 423                 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
 424                   __PF_RES_FIELD };
 425         int rc;
 426
 427         if (!MACHINE_IS_VM || pfault_disable)
 428                 return -1;
 429         asm volatile(
 430                 "       diag    %1,%0,0x258\n"
 431                 "0:     j       2f\n"
 432                 "1:     la      %0,8\n"
 433                 "2:\n"
 434                 EX_TABLE(0b,1b)
 435                 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
 436         __ctl_set_bit(0, 9);
 437         return rc;
 438 }
 439
 440 void pfault_fini(void)
 441 {
 442         pfault_refbk_t refbk =
 443         { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
 444
 445         if (!MACHINE_IS_VM || pfault_disable)
 446                 return;
 447         __ctl_clear_bit(0,9);
 448         asm volatile(
 449                 "       diag    %0,0,0x258\n"
 450                 "0:\n"
 451                 EX_TABLE(0b,0b)
 452                 : : "a" (&refbk), "m" (refbk) : "cc");
 453 }
 454
 455 static void pfault_interrupt(__u16 error_code)
 456 {
 457         struct task_struct *tsk;
 458         __u16 subcode;
 459
 460         /*
 461          * Get the external interruption subcode & pfault
 462          * initial/completion signal bit. VM stores this
 463          * in the 'cpu address' field associated with the
 464          * external interrupt.
 465          */
 466         subcode = S390_lowcore.cpu_addr;
 467         if ((subcode & 0xff00) != __SUBCODE_MASK)
 468                 return;
 469
 470         /*
 471          * Get the token (= address of the task structure of the affected task).
 472          */
 473         tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
 474
 475         if (subcode & 0x0080) {
 476                 /* signal bit is set -> a page has been swapped in by VM */
 477                 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
 478                         /* Initial interrupt was faster than the completion
 479                          * interrupt. pfault_wait is valid. Set pfault_wait
 480                          * back to zero and wake up the process. This can
 481                          * safely be done because the task is still sleeping
 482                          * and can't produce new pfaults. */
 483                         tsk->thread.pfault_wait = 0;
 484                         wake_up_process(tsk);
 485                         put_task_struct(tsk);
 486                 }
 487         } else {
 488                 /* signal bit not set -> a real page is missing. */
 489                 get_task_struct(tsk);
 490                 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 491                 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
 492                         /* Completion interrupt was faster than the initial
 493                          * interrupt (swapped in a -1 for pfault_wait). Set
 494                          * pfault_wait back to zero and exit. This can be
 495                          * done safely because tsk is running in kernel
 496                          * mode and can't produce new pfaults. */
 497                         tsk->thread.pfault_wait = 0;
 498                         set_task_state(tsk, TASK_RUNNING);
 499                         put_task_struct(tsk);
 500                 } else
 501                         set_tsk_need_resched(tsk);
 502         }
 503 }
 504
 505 void __init pfault_irq_init(void)
 506 {
 507         if (!MACHINE_IS_VM)
 508                 return;
 509
 510         /*
 511          * Try to get pfault pseudo page faults going.
 512          */
 513         if (register_early_external_interrupt(0x2603, pfault_interrupt,
 514                                               &ext_int_pfault) != 0)
 515                 panic("Couldn't request external interrupt 0x2603");
 516
 517         if (pfault_init() == 0)
 518                 return;
 519
 520         /* Tough luck, no pfault. */
 521         pfault_disable = 1;
 522         unregister_early_external_interrupt(0x2603, pfault_interrupt,
 523                                             &ext_int_pfault);
 524 }
 525 #endif