arch/powerpc/mm/fault.c

   1 /*
   2  *  PowerPC version
   3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   4  *
   5  *  Derived from "arch/i386/mm/fault.c"
   6  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   7  *
   8  *  Modified by Cort Dougan and Paul Mackerras.
   9  *
  10  *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
  11  *
  12  *  This program is free software; you can redistribute it and/or
  13  *  modify it under the terms of the GNU General Public License
  14  *  as published by the Free Software Foundation; either version
  15  *  2 of the License, or (at your option) any later version.
  16  */
  17
  18 #include <linux/signal.h>
  19 #include <linux/sched.h>
  20 #include <linux/kernel.h>
  21 #include <linux/errno.h>
  22 #include <linux/string.h>
  23 #include <linux/types.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/mman.h>
  26 #include <linux/mm.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/highmem.h>
  29 #include <linux/module.h>
  30 #include <linux/kprobes.h>
  31 #include <linux/kdebug.h>
  32 #include <linux/perf_event.h>
  33 #include <linux/magic.h>
  34 #include <linux/ratelimit.h>
  35
  36 #include <asm/firmware.h>
  37 #include <asm/page.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/mmu.h>
  40 #include <asm/mmu_context.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/siginfo.h>
  44 #include <asm/debug.h>
  45 #include <mm/mmu_decl.h>
  46
  47 #include "icswx.h"
  48
  49 #ifdef CONFIG_KPROBES
  50 static inline int notify_page_fault(struct pt_regs *regs)
  51 {
  52         int ret = 0;
  53
  54         /* kprobe_running() needs smp_processor_id() */
  55         if (!user_mode(regs)) {
  56                 preempt_disable();
  57                 if (kprobe_running() && kprobe_fault_handler(regs, 11))
  58                         ret = 1;
  59                 preempt_enable();
  60         }
  61
  62         return ret;
  63 }
  64 #else
  65 static inline int notify_page_fault(struct pt_regs *regs)
  66 {
  67         return 0;
  68 }
  69 #endif
  70
  71 /*
  72  * Check whether the instruction at regs->nip is a store using
  73  * an update addressing form which will update r1.
  74  */
  75 static int store_updates_sp(struct pt_regs *regs)
  76 {
  77         unsigned int inst;
  78
  79         if (get_user(inst, (unsigned int __user *)regs->nip))
  80                 return 0;
  81         /* check for 1 in the rA field */
  82         if (((inst >> 16) & 0x1f) != 1)
  83                 return 0;
  84         /* check major opcode */
  85         switch (inst >> 26) {
  86         case 37:        /* stwu */
  87         case 39:        /* stbu */
  88         case 45:        /* sthu */
  89         case 53:        /* stfsu */
  90         case 55:        /* stfdu */
  91                 return 1;
  92         case 62:        /* std or stdu */
  93                 return (inst & 3) == 1;
  94         case 31:
  95                 /* check minor opcode */
  96                 switch ((inst >> 1) & 0x3ff) {
  97                 case 181:       /* stdux */
  98                 case 183:       /* stwux */
  99                 case 247:       /* stbux */
 100                 case 439:       /* sthux */
 101                 case 695:       /* stfsux */
 102                 case 759:       /* stfdux */
 103                         return 1;
 104                 }
 105         }
 106         return 0;
 107 }
 108 /*
 109  * do_page_fault error handling helpers
 110  */
 111
 112 #define MM_FAULT_RETURN         0
 113 #define MM_FAULT_CONTINUE       -1
 114 #define MM_FAULT_ERR(sig)       (sig)
 115
 116 static int do_sigbus(struct pt_regs *regs, unsigned long address)
 117 {
 118         siginfo_t info;
 119
 120         up_read(&current->mm->mmap_sem);
 121
 122         if (user_mode(regs)) {
 123                 current->thread.trap_nr = BUS_ADRERR;
 124                 info.si_signo = SIGBUS;
 125                 info.si_errno = 0;
 126                 info.si_code = BUS_ADRERR;
 127                 info.si_addr = (void __user *)address;
 128                 force_sig_info(SIGBUS, &info, current);
 129                 return MM_FAULT_RETURN;
 130         }
 131         return MM_FAULT_ERR(SIGBUS);
 132 }
 133
 134 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
 135 {
 136         /*
 137          * Pagefault was interrupted by SIGKILL. We have no reason to
 138          * continue the pagefault.
 139          */
 140         if (fatal_signal_pending(current)) {
 141                 /*
 142                  * If we have retry set, the mmap semaphore will have
 143                  * alrady been released in __lock_page_or_retry(). Else
 144                  * we release it now.
 145                  */
 146                 if (!(fault & VM_FAULT_RETRY))
 147                         up_read(&current->mm->mmap_sem);
 148                 /* Coming from kernel, we need to deal with uaccess fixups */
 149                 if (user_mode(regs))
 150                         return MM_FAULT_RETURN;
 151                 return MM_FAULT_ERR(SIGKILL);
 152         }
 153
 154         /* No fault: be happy */
 155         if (!(fault & VM_FAULT_ERROR))
 156                 return MM_FAULT_CONTINUE;
 157
 158         /* Out of memory */
 159         if (fault & VM_FAULT_OOM) {
 160                 up_read(&current->mm->mmap_sem);
 161
 162                 /*
 163                  * We ran out of memory, or some other thing happened to us that
 164                  * made us unable to handle the page fault gracefully.
 165                  */
 166                 if (!user_mode(regs))
 167                         return MM_FAULT_ERR(SIGKILL);
 168                 pagefault_out_of_memory();
 169                 return MM_FAULT_RETURN;
 170         }
 171
 172         /* Bus error. x86 handles HWPOISON here, we'll add this if/when
 173          * we support the feature in HW
 174          */
 175         if (fault & VM_FAULT_SIGBUS)
 176                 return do_sigbus(regs, addr);
 177
 178         /* We don't understand the fault code, this is fatal */
 179         BUG();
 180         return MM_FAULT_CONTINUE;
 181 }
 182
 183 /*
 184  * For 600- and 800-family processors, the error_code parameter is DSISR
 185  * for a data fault, SRR1 for an instruction fault. For 400-family processors
 186  * the error_code parameter is ESR for a data fault, 0 for an instruction
 187  * fault.
 188  * For 64-bit processors, the error_code parameter is
 189  *  - DSISR for a non-SLB data access fault,
 190  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
 191  *  - 0 any SLB fault.
 192  *
 193  * The return value is 0 if the fault was handled, or the signal
 194  * number if this is a kernel fault that can't be handled here.
 195  */
 196 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 197                             unsigned long error_code)
 198 {
 199         struct vm_area_struct * vma;
 200         struct mm_struct *mm = current->mm;
 201         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 202         int code = SEGV_MAPERR;
 203         int is_write = 0;
 204         int trap = TRAP(regs);
 205         int is_exec = trap == 0x400;
 206         int fault;
 207
 208 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 209         /*
 210          * Fortunately the bit assignments in SRR1 for an instruction
 211          * fault and DSISR for a data fault are mostly the same for the
 212          * bits we are interested in.  But there are some bits which
 213          * indicate errors in DSISR but can validly be set in SRR1.
 214          */
 215         if (trap == 0x400)
 216                 error_code &= 0x48200000;
 217         else
 218                 is_write = error_code & DSISR_ISSTORE;
 219 #else
 220         is_write = error_code & ESR_DST;
 221 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 222
 223         if (is_write)
 224                 flags |= FAULT_FLAG_WRITE;
 225
 226 #ifdef CONFIG_PPC_ICSWX
 227         /*
 228          * we need to do this early because this "data storage
 229          * interrupt" does not update the DAR/DEAR so we don't want to
 230          * look at it
 231          */
 232         if (error_code & ICSWX_DSI_UCT) {
 233                 int rc = acop_handle_fault(regs, address, error_code);
 234                 if (rc)
 235                         return rc;
 236         }
 237 #endif /* CONFIG_PPC_ICSWX */
 238
 239         if (notify_page_fault(regs))
 240                 return 0;
 241
 242         if (unlikely(debugger_fault_handler(regs)))
 243                 return 0;
 244
 245         /* On a kernel SLB miss we can only check for a valid exception entry */
 246         if (!user_mode(regs) && (address >= TASK_SIZE))
 247                 return SIGSEGV;
 248
 249 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
 250                              defined(CONFIG_PPC_BOOK3S_64))
 251         if (error_code & DSISR_DABRMATCH) {
 252                 /* breakpoint match */
 253                 do_break(regs, address, error_code);
 254                 return 0;
 255         }
 256 #endif
 257
 258         /* We restore the interrupt state now */
 259         if (!arch_irq_disabled_regs(regs))
 260                 local_irq_enable();
 261
 262         if (in_atomic() || mm == NULL) {
 263                 if (!user_mode(regs))
 264                         return SIGSEGV;
 265                 /* in_atomic() in user mode is really bad,
 266                    as is current->mm == NULL. */
 267                 printk(KERN_EMERG "Page fault in user mode with "
 268                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 269                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 270                        regs->nip, regs->msr);
 271                 die("Weird page fault", regs, SIGSEGV);
 272         }
 273
 274         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 275
 276         /* When running in the kernel we expect faults to occur only to
 277          * addresses in user space.  All other faults represent errors in the
 278          * kernel and should generate an OOPS.  Unfortunately, in the case of an
 279          * erroneous fault occurring in a code path which already holds mmap_sem
 280          * we will deadlock attempting to validate the fault against the
 281          * address space.  Luckily the kernel only validly references user
 282          * space from well defined areas of code, which are listed in the
 283          * exceptions table.
 284          *
 285          * As the vast majority of faults will be valid we will only perform
 286          * the source reference check when there is a possibility of a deadlock.
 287          * Attempt to lock the address space, if we cannot we then validate the
 288          * source.  If this is invalid we can skip the address space check,
 289          * thus avoiding the deadlock.
 290          */
 291         if (!down_read_trylock(&mm->mmap_sem)) {
 292                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 293                         goto bad_area_nosemaphore;
 294
 295 retry:
 296                 down_read(&mm->mmap_sem);
 297         } else {
 298                 /*
 299                  * The above down_read_trylock() might have succeeded in
 300                  * which case we'll have missed the might_sleep() from
 301                  * down_read():
 302                  */
 303                 might_sleep();
 304         }
 305
 306         vma = find_vma(mm, address);
 307         if (!vma)
 308                 goto bad_area;
 309         if (vma->vm_start <= address)
 310                 goto good_area;
 311         if (!(vma->vm_flags & VM_GROWSDOWN))
 312                 goto bad_area;
 313
 314         /*
 315          * N.B. The POWER/Open ABI allows programs to access up to
 316          * 288 bytes below the stack pointer.
 317          * The kernel signal delivery code writes up to about 1.5kB
 318          * below the stack pointer (r1) before decrementing it.
 319          * The exec code can write slightly over 640kB to the stack
 320          * before setting the user r1.  Thus we allow the stack to
 321          * expand to 1MB without further checks.
 322          */
 323         if (address + 0x100000 < vma->vm_end) {
 324                 /* get user regs even if this fault is in kernel mode */
 325                 struct pt_regs *uregs = current->thread.regs;
 326                 if (uregs == NULL)
 327                         goto bad_area;
 328
 329                 /*
 330                  * A user-mode access to an address a long way below
 331                  * the stack pointer is only valid if the instruction
 332                  * is one which would update the stack pointer to the
 333                  * address accessed if the instruction completed,
 334                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 335                  * (or the byte, halfword, float or double forms).
 336                  *
 337                  * If we don't check this then any write to the area
 338                  * between the last mapped region and the stack will
 339                  * expand the stack rather than segfaulting.
 340                  */
 341                 if (address + 2048 < uregs->gpr[1]
 342                     && (!user_mode(regs) || !store_updates_sp(regs)))
 343                         goto bad_area;
 344         }
 345         if (expand_stack(vma, address))
 346                 goto bad_area;
 347
 348 good_area:
 349         code = SEGV_ACCERR;
 350 #if defined(CONFIG_6xx)
 351         if (error_code & 0x95700000)
 352                 /* an error such as lwarx to I/O controller space,
 353                    address matching DABR, eciwx, etc. */
 354                 goto bad_area;
 355 #endif /* CONFIG_6xx */
 356 #if defined(CONFIG_8xx)
 357         /* 8xx sometimes need to load a invalid/non-present TLBs.
 358          * These must be invalidated separately as linux mm don't.
 359          */
 360         if (error_code & 0x40000000) /* no translation? */
 361                 _tlbil_va(address, 0, 0, 0);
 362
 363         /* The MPC8xx seems to always set 0x80000000, which is
 364          * "undefined".  Of those that can be set, this is the only
 365          * one which seems bad.
 366          */
 367         if (error_code & 0x10000000)
 368                 /* Guarded storage error. */
 369                 goto bad_area;
 370 #endif /* CONFIG_8xx */
 371
 372         if (is_exec) {
 373 #ifdef CONFIG_PPC_STD_MMU
 374                 /* Protection fault on exec go straight to failure on
 375                  * Hash based MMUs as they either don't support per-page
 376                  * execute permission, or if they do, it's handled already
 377                  * at the hash level. This test would probably have to
 378                  * be removed if we change the way this works to make hash
 379                  * processors use the same I/D cache coherency mechanism
 380                  * as embedded.
 381                  */
 382                 if (error_code & DSISR_PROTFAULT)
 383                         goto bad_area;
 384 #endif /* CONFIG_PPC_STD_MMU */
 385
 386                 /*
 387                  * Allow execution from readable areas if the MMU does not
 388                  * provide separate controls over reading and executing.
 389                  *
 390                  * Note: That code used to not be enabled for 4xx/BookE.
 391                  * It is now as I/D cache coherency for these is done at
 392                  * set_pte_at() time and I see no reason why the test
 393                  * below wouldn't be valid on those processors. This -may-
 394                  * break programs compiled with a really old ABI though.
 395                  */
 396                 if (!(vma->vm_flags & VM_EXEC) &&
 397                     (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
 398                      !(vma->vm_flags & (VM_READ | VM_WRITE))))
 399                         goto bad_area;
 400         /* a write */
 401         } else if (is_write) {
 402                 if (!(vma->vm_flags & VM_WRITE))
 403                         goto bad_area;
 404         /* a read */
 405         } else {
 406                 /* protection fault */
 407                 if (error_code & 0x08000000)
 408                         goto bad_area;
 409                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 410                         goto bad_area;
 411         }
 412
 413         /*
 414          * If for any reason at all we couldn't handle the fault,
 415          * make sure we exit gracefully rather than endlessly redo
 416          * the fault.
 417          */
 418         fault = handle_mm_fault(mm, vma, address, flags);
 419         if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
 420                 int rc = mm_fault_error(regs, address, fault);
 421                 if (rc >= MM_FAULT_RETURN)
 422                         return rc;
 423         }
 424
 425         /*
 426          * Major/minor page fault accounting is only done on the
 427          * initial attempt. If we go through a retry, it is extremely
 428          * likely that the page will be found in page cache at that point.
 429          */
 430         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 431                 if (fault & VM_FAULT_MAJOR) {
 432                         current->maj_flt++;
 433                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 434                                       regs, address);
 435 #ifdef CONFIG_PPC_SMLPAR
 436                         if (firmware_has_feature(FW_FEATURE_CMO)) {
 437                                 preempt_disable();
 438                                 get_lppaca()->page_ins += (1 << PAGE_FACTOR);
 439                                 preempt_enable();
 440                         }
 441 #endif /* CONFIG_PPC_SMLPAR */
 442                 } else {
 443                         current->min_flt++;
 444                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 445                                       regs, address);
 446                 }
 447                 if (fault & VM_FAULT_RETRY) {
 448                         /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
 449                          * of starvation. */
 450                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 451                         flags |= FAULT_FLAG_TRIED;
 452                         goto retry;
 453                 }
 454         }
 455
 456         up_read(&mm->mmap_sem);
 457         return 0;
 458
 459 bad_area:
 460         up_read(&mm->mmap_sem);
 461
 462 bad_area_nosemaphore:
 463         /* User mode accesses cause a SIGSEGV */
 464         if (user_mode(regs)) {
 465                 _exception(SIGSEGV, regs, code, address);
 466                 return 0;
 467         }
 468
 469         if (is_exec && (error_code & DSISR_PROTFAULT))
 470                 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
 471                                    " page (%lx) - exploit attempt? (uid: %d)\n",
 472                                    address, from_kuid(&init_user_ns, current_uid()));
 473
 474         return SIGSEGV;
 475
 476 }
 477
 478 /*
 479  * bad_page_fault is called when we have a bad access from the kernel.
 480  * It is called from the DSI and ISI handlers in head.S and from some
 481  * of the procedures in traps.c.
 482  */
 483 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 484 {
 485         const struct exception_table_entry *entry;
 486         unsigned long *stackend;
 487
 488         /* Are we prepared to handle this fault?  */
 489         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 490                 regs->nip = entry->fixup;
 491                 return;
 492         }
 493
 494         /* kernel has accessed a bad area */
 495
 496         switch (regs->trap) {
 497         case 0x300:
 498         case 0x380:
 499                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 500                         "data at address 0x%08lx\n", regs->dar);
 501                 break;
 502         case 0x400:
 503         case 0x480:
 504                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 505                         "instruction fetch\n");
 506                 break;
 507         default:
 508                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 509                         "unknown fault\n");
 510                 break;
 511         }
 512         printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
 513                 regs->nip);
 514
 515         stackend = end_of_stack(current);
 516         if (current != &init_task && *stackend != STACK_END_MAGIC)
 517                 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
 518
 519         die("Kernel access of bad area", regs, sig);
 520 }