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 #include <linux/context_tracking.h>
  36
  37 #include <asm/firmware.h>
  38 #include <asm/page.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/mmu.h>
  41 #include <asm/mmu_context.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/tlbflush.h>
  44 #include <asm/siginfo.h>
  45 #include <asm/debug.h>
  46 #include <mm/mmu_decl.h>
  47
  48 #include "icswx.h"
  49
  50 #ifdef CONFIG_KPROBES
  51 static inline int notify_page_fault(struct pt_regs *regs)
  52 {
  53         int ret = 0;
  54
  55         /* kprobe_running() needs smp_processor_id() */
  56         if (!user_mode(regs)) {
  57                 preempt_disable();
  58                 if (kprobe_running() && kprobe_fault_handler(regs, 11))
  59                         ret = 1;
  60                 preempt_enable();
  61         }
  62
  63         return ret;
  64 }
  65 #else
  66 static inline int notify_page_fault(struct pt_regs *regs)
  67 {
  68         return 0;
  69 }
  70 #endif
  71
  72 /*
  73  * Check whether the instruction at regs->nip is a store using
  74  * an update addressing form which will update r1.
  75  */
  76 static int store_updates_sp(struct pt_regs *regs)
  77 {
  78         unsigned int inst;
  79
  80         if (get_user(inst, (unsigned int __user *)regs->nip))
  81                 return 0;
  82         /* check for 1 in the rA field */
  83         if (((inst >> 16) & 0x1f) != 1)
  84                 return 0;
  85         /* check major opcode */
  86         switch (inst >> 26) {
  87         case 37:        /* stwu */
  88         case 39:        /* stbu */
  89         case 45:        /* sthu */
  90         case 53:        /* stfsu */
  91         case 55:        /* stfdu */
  92                 return 1;
  93         case 62:        /* std or stdu */
  94                 return (inst & 3) == 1;
  95         case 31:
  96                 /* check minor opcode */
  97                 switch ((inst >> 1) & 0x3ff) {
  98                 case 181:       /* stdux */
  99                 case 183:       /* stwux */
 100                 case 247:       /* stbux */
 101                 case 439:       /* sthux */
 102                 case 695:       /* stfsux */
 103                 case 759:       /* stfdux */
 104                         return 1;
 105                 }
 106         }
 107         return 0;
 108 }
 109 /*
 110  * do_page_fault error handling helpers
 111  */
 112
 113 #define MM_FAULT_RETURN         0
 114 #define MM_FAULT_CONTINUE       -1
 115 #define MM_FAULT_ERR(sig)       (sig)
 116
 117 static int do_sigbus(struct pt_regs *regs, unsigned long address)
 118 {
 119         siginfo_t info;
 120
 121         up_read(&current->mm->mmap_sem);
 122
 123         if (user_mode(regs)) {
 124                 current->thread.trap_nr = BUS_ADRERR;
 125                 info.si_signo = SIGBUS;
 126                 info.si_errno = 0;
 127                 info.si_code = BUS_ADRERR;
 128                 info.si_addr = (void __user *)address;
 129                 force_sig_info(SIGBUS, &info, current);
 130                 return MM_FAULT_RETURN;
 131         }
 132         return MM_FAULT_ERR(SIGBUS);
 133 }
 134
 135 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
 136 {
 137         /*
 138          * Pagefault was interrupted by SIGKILL. We have no reason to
 139          * continue the pagefault.
 140          */
 141         if (fatal_signal_pending(current)) {
 142                 /*
 143                  * If we have retry set, the mmap semaphore will have
 144                  * alrady been released in __lock_page_or_retry(). Else
 145                  * we release it now.
 146                  */
 147                 if (!(fault & VM_FAULT_RETRY))
 148                         up_read(&current->mm->mmap_sem);
 149                 /* Coming from kernel, we need to deal with uaccess fixups */
 150                 if (user_mode(regs))
 151                         return MM_FAULT_RETURN;
 152                 return MM_FAULT_ERR(SIGKILL);
 153         }
 154
 155         /* No fault: be happy */
 156         if (!(fault & VM_FAULT_ERROR))
 157                 return MM_FAULT_CONTINUE;
 158
 159         /* Out of memory */
 160         if (fault & VM_FAULT_OOM) {
 161                 up_read(&current->mm->mmap_sem);
 162
 163                 /*
 164                  * We ran out of memory, or some other thing happened to us that
 165                  * made us unable to handle the page fault gracefully.
 166                  */
 167                 if (!user_mode(regs))
 168                         return MM_FAULT_ERR(SIGKILL);
 169                 pagefault_out_of_memory();
 170                 return MM_FAULT_RETURN;
 171         }
 172
 173         /* Bus error. x86 handles HWPOISON here, we'll add this if/when
 174          * we support the feature in HW
 175          */
 176         if (fault & VM_FAULT_SIGBUS)
 177                 return do_sigbus(regs, addr);
 178
 179         /* We don't understand the fault code, this is fatal */
 180         BUG();
 181         return MM_FAULT_CONTINUE;
 182 }
 183
 184 /*
 185  * For 600- and 800-family processors, the error_code parameter is DSISR
 186  * for a data fault, SRR1 for an instruction fault. For 400-family processors
 187  * the error_code parameter is ESR for a data fault, 0 for an instruction
 188  * fault.
 189  * For 64-bit processors, the error_code parameter is
 190  *  - DSISR for a non-SLB data access fault,
 191  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
 192  *  - 0 any SLB fault.
 193  *
 194  * The return value is 0 if the fault was handled, or the signal
 195  * number if this is a kernel fault that can't be handled here.
 196  */
 197 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 198                             unsigned long error_code)
 199 {
 200         enum ctx_state prev_state = exception_enter();
 201         struct vm_area_struct * vma;
 202         struct mm_struct *mm = current->mm;
 203         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 204         int code = SEGV_MAPERR;
 205         int is_write = 0;
 206         int trap = TRAP(regs);
 207         int is_exec = trap == 0x400;
 208         int fault;
 209         int rc = 0;
 210
 211 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 212         /*
 213          * Fortunately the bit assignments in SRR1 for an instruction
 214          * fault and DSISR for a data fault are mostly the same for the
 215          * bits we are interested in.  But there are some bits which
 216          * indicate errors in DSISR but can validly be set in SRR1.
 217          */
 218         if (trap == 0x400)
 219                 error_code &= 0x48200000;
 220         else
 221                 is_write = error_code & DSISR_ISSTORE;
 222 #else
 223         is_write = error_code & ESR_DST;
 224 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 225
 226         if (is_write)
 227                 flags |= FAULT_FLAG_WRITE;
 228
 229 #ifdef CONFIG_PPC_ICSWX
 230         /*
 231          * we need to do this early because this "data storage
 232          * interrupt" does not update the DAR/DEAR so we don't want to
 233          * look at it
 234          */
 235         if (error_code & ICSWX_DSI_UCT) {
 236                 rc = acop_handle_fault(regs, address, error_code);
 237                 if (rc)
 238                         goto bail;
 239         }
 240 #endif /* CONFIG_PPC_ICSWX */
 241
 242         if (notify_page_fault(regs))
 243                 goto bail;
 244
 245         if (unlikely(debugger_fault_handler(regs)))
 246                 goto bail;
 247
 248         /* On a kernel SLB miss we can only check for a valid exception entry */
 249         if (!user_mode(regs) && (address >= TASK_SIZE)) {
 250                 rc = SIGSEGV;
 251                 goto bail;
 252         }
 253
 254 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
 255                              defined(CONFIG_PPC_BOOK3S_64))
 256         if (error_code & DSISR_DABRMATCH) {
 257                 /* breakpoint match */
 258                 do_break(regs, address, error_code);
 259                 goto bail;
 260         }
 261 #endif
 262
 263         /* We restore the interrupt state now */
 264         if (!arch_irq_disabled_regs(regs))
 265                 local_irq_enable();
 266
 267         if (in_atomic() || mm == NULL) {
 268                 if (!user_mode(regs)) {
 269                         rc = SIGSEGV;
 270                         goto bail;
 271                 }
 272                 /* in_atomic() in user mode is really bad,
 273                    as is current->mm == NULL. */
 274                 printk(KERN_EMERG "Page fault in user mode with "
 275                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 276                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 277                        regs->nip, regs->msr);
 278                 die("Weird page fault", regs, SIGSEGV);
 279         }
 280
 281         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 282
 283         /* When running in the kernel we expect faults to occur only to
 284          * addresses in user space.  All other faults represent errors in the
 285          * kernel and should generate an OOPS.  Unfortunately, in the case of an
 286          * erroneous fault occurring in a code path which already holds mmap_sem
 287          * we will deadlock attempting to validate the fault against the
 288          * address space.  Luckily the kernel only validly references user
 289          * space from well defined areas of code, which are listed in the
 290          * exceptions table.
 291          *
 292          * As the vast majority of faults will be valid we will only perform
 293          * the source reference check when there is a possibility of a deadlock.
 294          * Attempt to lock the address space, if we cannot we then validate the
 295          * source.  If this is invalid we can skip the address space check,
 296          * thus avoiding the deadlock.
 297          */
 298         if (!down_read_trylock(&mm->mmap_sem)) {
 299                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 300                         goto bad_area_nosemaphore;
 301
 302 retry:
 303                 down_read(&mm->mmap_sem);
 304         } else {
 305                 /*
 306                  * The above down_read_trylock() might have succeeded in
 307                  * which case we'll have missed the might_sleep() from
 308                  * down_read():
 309                  */
 310                 might_sleep();
 311         }
 312
 313         vma = find_vma(mm, address);
 314         if (!vma)
 315                 goto bad_area;
 316         if (vma->vm_start <= address)
 317                 goto good_area;
 318         if (!(vma->vm_flags & VM_GROWSDOWN))
 319                 goto bad_area;
 320
 321         /*
 322          * N.B. The POWER/Open ABI allows programs to access up to
 323          * 288 bytes below the stack pointer.
 324          * The kernel signal delivery code writes up to about 1.5kB
 325          * below the stack pointer (r1) before decrementing it.
 326          * The exec code can write slightly over 640kB to the stack
 327          * before setting the user r1.  Thus we allow the stack to
 328          * expand to 1MB without further checks.
 329          */
 330         if (address + 0x100000 < vma->vm_end) {
 331                 /* get user regs even if this fault is in kernel mode */
 332                 struct pt_regs *uregs = current->thread.regs;
 333                 if (uregs == NULL)
 334                         goto bad_area;
 335
 336                 /*
 337                  * A user-mode access to an address a long way below
 338                  * the stack pointer is only valid if the instruction
 339                  * is one which would update the stack pointer to the
 340                  * address accessed if the instruction completed,
 341                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 342                  * (or the byte, halfword, float or double forms).
 343                  *
 344                  * If we don't check this then any write to the area
 345                  * between the last mapped region and the stack will
 346                  * expand the stack rather than segfaulting.
 347                  */
 348                 if (address + 2048 < uregs->gpr[1]
 349                     && (!user_mode(regs) || !store_updates_sp(regs)))
 350                         goto bad_area;
 351         }
 352         if (expand_stack(vma, address))
 353                 goto bad_area;
 354
 355 good_area:
 356         code = SEGV_ACCERR;
 357 #if defined(CONFIG_6xx)
 358         if (error_code & 0x95700000)
 359                 /* an error such as lwarx to I/O controller space,
 360                    address matching DABR, eciwx, etc. */
 361                 goto bad_area;
 362 #endif /* CONFIG_6xx */
 363 #if defined(CONFIG_8xx)
 364         /* 8xx sometimes need to load a invalid/non-present TLBs.
 365          * These must be invalidated separately as linux mm don't.
 366          */
 367         if (error_code & 0x40000000) /* no translation? */
 368                 _tlbil_va(address, 0, 0, 0);
 369
 370         /* The MPC8xx seems to always set 0x80000000, which is
 371          * "undefined".  Of those that can be set, this is the only
 372          * one which seems bad.
 373          */
 374         if (error_code & 0x10000000)
 375                 /* Guarded storage error. */
 376                 goto bad_area;
 377 #endif /* CONFIG_8xx */
 378
 379         if (is_exec) {
 380 #ifdef CONFIG_PPC_STD_MMU
 381                 /* Protection fault on exec go straight to failure on
 382                  * Hash based MMUs as they either don't support per-page
 383                  * execute permission, or if they do, it's handled already
 384                  * at the hash level. This test would probably have to
 385                  * be removed if we change the way this works to make hash
 386                  * processors use the same I/D cache coherency mechanism
 387                  * as embedded.
 388                  */
 389                 if (error_code & DSISR_PROTFAULT)
 390                         goto bad_area;
 391 #endif /* CONFIG_PPC_STD_MMU */
 392
 393                 /*
 394                  * Allow execution from readable areas if the MMU does not
 395                  * provide separate controls over reading and executing.
 396                  *
 397                  * Note: That code used to not be enabled for 4xx/BookE.
 398                  * It is now as I/D cache coherency for these is done at
 399                  * set_pte_at() time and I see no reason why the test
 400                  * below wouldn't be valid on those processors. This -may-
 401                  * break programs compiled with a really old ABI though.
 402                  */
 403                 if (!(vma->vm_flags & VM_EXEC) &&
 404                     (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
 405                      !(vma->vm_flags & (VM_READ | VM_WRITE))))
 406                         goto bad_area;
 407         /* a write */
 408         } else if (is_write) {
 409                 if (!(vma->vm_flags & VM_WRITE))
 410                         goto bad_area;
 411         /* a read */
 412         } else {
 413                 /* protection fault */
 414                 if (error_code & 0x08000000)
 415                         goto bad_area;
 416                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 417                         goto bad_area;
 418         }
 419
 420         /*
 421          * If for any reason at all we couldn't handle the fault,
 422          * make sure we exit gracefully rather than endlessly redo
 423          * the fault.
 424          */
 425         fault = handle_mm_fault(mm, vma, address, flags);
 426         if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
 427                 rc = mm_fault_error(regs, address, fault);
 428                 if (rc >= MM_FAULT_RETURN)
 429                         goto bail;
 430                 else
 431                         rc = 0;
 432         }
 433
 434         /*
 435          * Major/minor page fault accounting is only done on the
 436          * initial attempt. If we go through a retry, it is extremely
 437          * likely that the page will be found in page cache at that point.
 438          */
 439         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 440                 if (fault & VM_FAULT_MAJOR) {
 441                         current->maj_flt++;
 442                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 443                                       regs, address);
 444 #ifdef CONFIG_PPC_SMLPAR
 445                         if (firmware_has_feature(FW_FEATURE_CMO)) {
 446                                 preempt_disable();
 447                                 get_lppaca()->page_ins += (1 << PAGE_FACTOR);
 448                                 preempt_enable();
 449                         }
 450 #endif /* CONFIG_PPC_SMLPAR */
 451                 } else {
 452                         current->min_flt++;
 453                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 454                                       regs, address);
 455                 }
 456                 if (fault & VM_FAULT_RETRY) {
 457                         /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
 458                          * of starvation. */
 459                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 460                         flags |= FAULT_FLAG_TRIED;
 461                         goto retry;
 462                 }
 463         }
 464
 465         up_read(&mm->mmap_sem);
 466         goto bail;
 467
 468 bad_area:
 469         up_read(&mm->mmap_sem);
 470
 471 bad_area_nosemaphore:
 472         /* User mode accesses cause a SIGSEGV */
 473         if (user_mode(regs)) {
 474                 _exception(SIGSEGV, regs, code, address);
 475                 goto bail;
 476         }
 477
 478         if (is_exec && (error_code & DSISR_PROTFAULT))
 479                 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
 480                                    " page (%lx) - exploit attempt? (uid: %d)\n",
 481                                    address, from_kuid(&init_user_ns, current_uid()));
 482
 483         rc = SIGSEGV;
 484
 485 bail:
 486         exception_exit(prev_state);
 487         return rc;
 488
 489 }
 490
 491 /*
 492  * bad_page_fault is called when we have a bad access from the kernel.
 493  * It is called from the DSI and ISI handlers in head.S and from some
 494  * of the procedures in traps.c.
 495  */
 496 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 497 {
 498         const struct exception_table_entry *entry;
 499         unsigned long *stackend;
 500
 501         /* Are we prepared to handle this fault?  */
 502         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 503                 regs->nip = entry->fixup;
 504                 return;
 505         }
 506
 507         /* kernel has accessed a bad area */
 508
 509         switch (regs->trap) {
 510         case 0x300:
 511         case 0x380:
 512                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 513                         "data at address 0x%08lx\n", regs->dar);
 514                 break;
 515         case 0x400:
 516         case 0x480:
 517                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 518                         "instruction fetch\n");
 519                 break;
 520         default:
 521                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 522                         "unknown fault\n");
 523                 break;
 524         }
 525         printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
 526                 regs->nip);
 527
 528         stackend = end_of_stack(current);
 529         if (current != &init_task && *stackend != STACK_END_MAGIC)
 530                 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
 531
 532         die("Kernel access of bad area", regs, sig);
 533 }