arch/sparc/mm/fault_64.c

   1 /*
   2  * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
   3  *
   4  * Copyright (C) 1996, 2008 David S. Miller (davem@davemloft.net)
   5  * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
   6  */
   7
   8 #include <asm/head.h>
   9
  10 #include <linux/string.h>
  11 #include <linux/types.h>
  12 #include <linux/sched.h>
  13 #include <linux/sched/debug.h>
  14 #include <linux/ptrace.h>
  15 #include <linux/mman.h>
  16 #include <linux/signal.h>
  17 #include <linux/mm.h>
  18 #include <linux/extable.h>
  19 #include <linux/init.h>
  20 #include <linux/perf_event.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/kprobes.h>
  23 #include <linux/kdebug.h>
  24 #include <linux/percpu.h>
  25 #include <linux/context_tracking.h>
  26 #include <linux/uaccess.h>
  27
  28 #include <asm/page.h>
  29 #include <asm/pgtable.h>
  30 #include <asm/openprom.h>
  31 #include <asm/oplib.h>
  32 #include <asm/asi.h>
  33 #include <asm/lsu.h>
  34 #include <asm/sections.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/setup.h>
  37
  38 int show_unhandled_signals = 1;
  39
  40 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
  41 {
  42         int ret = 0;
  43
  44         /* kprobe_running() needs smp_processor_id() */
  45         if (kprobes_built_in() && !user_mode(regs)) {
  46                 preempt_disable();
  47                 if (kprobe_running() && kprobe_fault_handler(regs, 0))
  48                         ret = 1;
  49                 preempt_enable();
  50         }
  51         return ret;
  52 }
  53
  54 static void __kprobes unhandled_fault(unsigned long address,
  55                                       struct task_struct *tsk,
  56                                       struct pt_regs *regs)
  57 {
  58         if ((unsigned long) address < PAGE_SIZE) {
  59                 printk(KERN_ALERT "Unable to handle kernel NULL "
  60                        "pointer dereference\n");
  61         } else {
  62                 printk(KERN_ALERT "Unable to handle kernel paging request "
  63                        "at virtual address %016lx\n", (unsigned long)address);
  64         }
  65         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  66                (tsk->mm ?
  67                 CTX_HWBITS(tsk->mm->context) :
  68                 CTX_HWBITS(tsk->active_mm->context)));
  69         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  70                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  71                           (unsigned long) tsk->active_mm->pgd));
  72         die_if_kernel("Oops", regs);
  73 }
  74
  75 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  76 {
  77         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  78                regs->tpc);
  79         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  80         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  81         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  82         dump_stack();
  83         unhandled_fault(regs->tpc, current, regs);
  84 }
  85
  86 /*
  87  * We now make sure that mmap_sem is held in all paths that call
  88  * this. Additionally, to prevent kswapd from ripping ptes from
  89  * under us, raise interrupts around the time that we look at the
  90  * pte, kswapd will have to wait to get his smp ipi response from
  91  * us. vmtruncate likewise. This saves us having to get pte lock.
  92  */
  93 static unsigned int get_user_insn(unsigned long tpc)
  94 {
  95         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  96         pud_t *pudp;
  97         pmd_t *pmdp;
  98         pte_t *ptep, pte;
  99         unsigned long pa;
 100         u32 insn = 0;
 101
 102         if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
 103                 goto out;
 104         pudp = pud_offset(pgdp, tpc);
 105         if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
 106                 goto out;
 107
 108         /* This disables preemption for us as well. */
 109         local_irq_disable();
 110
 111         pmdp = pmd_offset(pudp, tpc);
 112         if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
 113                 goto out_irq_enable;
 114
 115 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 116         if (is_hugetlb_pmd(*pmdp)) {
 117                 pa  = pmd_pfn(*pmdp) << PAGE_SHIFT;
 118                 pa += tpc & ~HPAGE_MASK;
 119
 120                 /* Use phys bypass so we don't pollute dtlb/dcache. */
 121                 __asm__ __volatile__("lduwa [%1] %2, %0"
 122                                      : "=r" (insn)
 123                                      : "r" (pa), "i" (ASI_PHYS_USE_EC));
 124         } else
 125 #endif
 126         {
 127                 ptep = pte_offset_map(pmdp, tpc);
 128                 pte = *ptep;
 129                 if (pte_present(pte)) {
 130                         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 131                         pa += (tpc & ~PAGE_MASK);
 132
 133                         /* Use phys bypass so we don't pollute dtlb/dcache. */
 134                         __asm__ __volatile__("lduwa [%1] %2, %0"
 135                                              : "=r" (insn)
 136                                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 137                 }
 138                 pte_unmap(ptep);
 139         }
 140 out_irq_enable:
 141         local_irq_enable();
 142 out:
 143         return insn;
 144 }
 145
 146 static inline void
 147 show_signal_msg(struct pt_regs *regs, int sig, int code,
 148                 unsigned long address, struct task_struct *tsk)
 149 {
 150         if (!unhandled_signal(tsk, sig))
 151                 return;
 152
 153         if (!printk_ratelimit())
 154                 return;
 155
 156         printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
 157                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 158                tsk->comm, task_pid_nr(tsk), address,
 159                (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
 160                (void *)regs->u_regs[UREG_FP], code);
 161
 162         print_vma_addr(KERN_CONT " in ", regs->tpc);
 163
 164         printk(KERN_CONT "\n");
 165 }
 166
 167 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 168                              unsigned long fault_addr, unsigned int insn,
 169                              int fault_code)
 170 {
 171         unsigned long addr;
 172         siginfo_t info;
 173
 174         info.si_code = code;
 175         info.si_signo = sig;
 176         info.si_errno = 0;
 177         if (fault_code & FAULT_CODE_ITLB) {
 178                 addr = regs->tpc;
 179         } else {
 180                 /* If we were able to probe the faulting instruction, use it
 181                  * to compute a precise fault address.  Otherwise use the fault
 182                  * time provided address which may only have page granularity.
 183                  */
 184                 if (insn)
 185                         addr = compute_effective_address(regs, insn, 0);
 186                 else
 187                         addr = fault_addr;
 188         }
 189         info.si_addr = (void __user *) addr;
 190         info.si_trapno = 0;
 191
 192         if (unlikely(show_unhandled_signals))
 193                 show_signal_msg(regs, sig, code, addr, current);
 194
 195         force_sig_info(sig, &info, current);
 196 }
 197
 198 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 199 {
 200         if (!insn) {
 201                 if (!regs->tpc || (regs->tpc & 0x3))
 202                         return 0;
 203                 if (regs->tstate & TSTATE_PRIV) {
 204                         insn = *(unsigned int *) regs->tpc;
 205                 } else {
 206                         insn = get_user_insn(regs->tpc);
 207                 }
 208         }
 209         return insn;
 210 }
 211
 212 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 213                                       int fault_code, unsigned int insn,
 214                                       unsigned long address)
 215 {
 216         unsigned char asi = ASI_P;
 217
 218         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 219                 goto cannot_handle;
 220
 221         /* If user insn could be read (thus insn is zero), that
 222          * is fine.  We will just gun down the process with a signal
 223          * in that case.
 224          */
 225
 226         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 227             (insn & 0xc0800000) == 0xc0800000) {
 228                 if (insn & 0x2000)
 229                         asi = (regs->tstate >> 24);
 230                 else
 231                         asi = (insn >> 5);
 232                 if ((asi & 0xf2) == 0x82) {
 233                         if (insn & 0x1000000) {
 234                                 handle_ldf_stq(insn, regs);
 235                         } else {
 236                                 /* This was a non-faulting load. Just clear the
 237                                  * destination register(s) and continue with the next
 238                                  * instruction. -jj
 239                                  */
 240                                 handle_ld_nf(insn, regs);
 241                         }
 242                         return;
 243                 }
 244         }
 245
 246         /* Is this in ex_table? */
 247         if (regs->tstate & TSTATE_PRIV) {
 248                 const struct exception_table_entry *entry;
 249
 250                 entry = search_exception_tables(regs->tpc);
 251                 if (entry) {
 252                         regs->tpc = entry->fixup;
 253                         regs->tnpc = regs->tpc + 4;
 254                         return;
 255                 }
 256         } else {
 257                 /* The si_code was set to make clear whether
 258                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 259                  */
 260                 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
 261                 return;
 262         }
 263
 264 cannot_handle:
 265         unhandled_fault (address, current, regs);
 266 }
 267
 268 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 269 {
 270         static int times;
 271
 272         if (times++ < 10)
 273                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 274                        "64-bit TPC [%lx]\n",
 275                        current->comm, current->pid,
 276                        regs->tpc);
 277         show_regs(regs);
 278 }
 279
 280 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 281 {
 282         enum ctx_state prev_state = exception_enter();
 283         struct mm_struct *mm = current->mm;
 284         struct vm_area_struct *vma;
 285         unsigned int insn = 0;
 286         int si_code, fault_code, fault;
 287         unsigned long address, mm_rss;
 288         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 289
 290         fault_code = get_thread_fault_code();
 291
 292         if (notify_page_fault(regs))
 293                 goto exit_exception;
 294
 295         si_code = SEGV_MAPERR;
 296         address = current_thread_info()->fault_address;
 297
 298         if ((fault_code & FAULT_CODE_ITLB) &&
 299             (fault_code & FAULT_CODE_DTLB))
 300                 BUG();
 301
 302         if (test_thread_flag(TIF_32BIT)) {
 303                 if (!(regs->tstate & TSTATE_PRIV)) {
 304                         if (unlikely((regs->tpc >> 32) != 0)) {
 305                                 bogus_32bit_fault_tpc(regs);
 306                                 goto intr_or_no_mm;
 307                         }
 308                 }
 309                 if (unlikely((address >> 32) != 0))
 310                         goto intr_or_no_mm;
 311         }
 312
 313         if (regs->tstate & TSTATE_PRIV) {
 314                 unsigned long tpc = regs->tpc;
 315
 316                 /* Sanity check the PC. */
 317                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 318                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 319                         /* Valid, no problems... */
 320                 } else {
 321                         bad_kernel_pc(regs, address);
 322                         goto exit_exception;
 323                 }
 324         } else
 325                 flags |= FAULT_FLAG_USER;
 326
 327         /*
 328          * If we're in an interrupt or have no user
 329          * context, we must not take the fault..
 330          */
 331         if (faulthandler_disabled() || !mm)
 332                 goto intr_or_no_mm;
 333
 334         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 335
 336         if (!down_read_trylock(&mm->mmap_sem)) {
 337                 if ((regs->tstate & TSTATE_PRIV) &&
 338                     !search_exception_tables(regs->tpc)) {
 339                         insn = get_fault_insn(regs, insn);
 340                         goto handle_kernel_fault;
 341                 }
 342
 343 retry:
 344                 down_read(&mm->mmap_sem);
 345         }
 346
 347         if (fault_code & FAULT_CODE_BAD_RA)
 348                 goto do_sigbus;
 349
 350         vma = find_vma(mm, address);
 351         if (!vma)
 352                 goto bad_area;
 353
 354         /* Pure DTLB misses do not tell us whether the fault causing
 355          * load/store/atomic was a write or not, it only says that there
 356          * was no match.  So in such a case we (carefully) read the
 357          * instruction to try and figure this out.  It's an optimization
 358          * so it's ok if we can't do this.
 359          *
 360          * Special hack, window spill/fill knows the exact fault type.
 361          */
 362         if (((fault_code &
 363               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 364             (vma->vm_flags & VM_WRITE) != 0) {
 365                 insn = get_fault_insn(regs, 0);
 366                 if (!insn)
 367                         goto continue_fault;
 368                 /* All loads, stores and atomics have bits 30 and 31 both set
 369                  * in the instruction.  Bit 21 is set in all stores, but we
 370                  * have to avoid prefetches which also have bit 21 set.
 371                  */
 372                 if ((insn & 0xc0200000) == 0xc0200000 &&
 373                     (insn & 0x01780000) != 0x01680000) {
 374                         /* Don't bother updating thread struct value,
 375                          * because update_mmu_cache only cares which tlb
 376                          * the access came from.
 377                          */
 378                         fault_code |= FAULT_CODE_WRITE;
 379                 }
 380         }
 381 continue_fault:
 382
 383         if (vma->vm_start <= address)
 384                 goto good_area;
 385         if (!(vma->vm_flags & VM_GROWSDOWN))
 386                 goto bad_area;
 387         if (!(fault_code & FAULT_CODE_WRITE)) {
 388                 /* Non-faulting loads shouldn't expand stack. */
 389                 insn = get_fault_insn(regs, insn);
 390                 if ((insn & 0xc0800000) == 0xc0800000) {
 391                         unsigned char asi;
 392
 393                         if (insn & 0x2000)
 394                                 asi = (regs->tstate >> 24);
 395                         else
 396                                 asi = (insn >> 5);
 397                         if ((asi & 0xf2) == 0x82)
 398                                 goto bad_area;
 399                 }
 400         }
 401         if (expand_stack(vma, address))
 402                 goto bad_area;
 403         /*
 404          * Ok, we have a good vm_area for this memory access, so
 405          * we can handle it..
 406          */
 407 good_area:
 408         si_code = SEGV_ACCERR;
 409
 410         /* If we took a ITLB miss on a non-executable page, catch
 411          * that here.
 412          */
 413         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 414                 WARN(address != regs->tpc,
 415                      "address (%lx) != regs->tpc (%lx)\n", address, regs->tpc);
 416                 WARN_ON(regs->tstate & TSTATE_PRIV);
 417                 goto bad_area;
 418         }
 419
 420         if (fault_code & FAULT_CODE_WRITE) {
 421                 if (!(vma->vm_flags & VM_WRITE))
 422                         goto bad_area;
 423
 424                 /* Spitfire has an icache which does not snoop
 425                  * processor stores.  Later processors do...
 426                  */
 427                 if (tlb_type == spitfire &&
 428                     (vma->vm_flags & VM_EXEC) != 0 &&
 429                     vma->vm_file != NULL)
 430                         set_thread_fault_code(fault_code |
 431                                               FAULT_CODE_BLKCOMMIT);
 432
 433                 flags |= FAULT_FLAG_WRITE;
 434         } else {
 435                 /* Allow reads even for write-only mappings */
 436                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 437                         goto bad_area;
 438         }
 439
 440         fault = handle_mm_fault(vma, address, flags);
 441
 442         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 443                 goto exit_exception;
 444
 445         if (unlikely(fault & VM_FAULT_ERROR)) {
 446                 if (fault & VM_FAULT_OOM)
 447                         goto out_of_memory;
 448                 else if (fault & VM_FAULT_SIGSEGV)
 449                         goto bad_area;
 450                 else if (fault & VM_FAULT_SIGBUS)
 451                         goto do_sigbus;
 452                 BUG();
 453         }
 454
 455         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 456                 if (fault & VM_FAULT_MAJOR) {
 457                         current->maj_flt++;
 458                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
 459                                       1, regs, address);
 460                 } else {
 461                         current->min_flt++;
 462                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
 463                                       1, regs, address);
 464                 }
 465                 if (fault & VM_FAULT_RETRY) {
 466                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 467                         flags |= FAULT_FLAG_TRIED;
 468
 469                         /* No need to up_read(&mm->mmap_sem) as we would
 470                          * have already released it in __lock_page_or_retry
 471                          * in mm/filemap.c.
 472                          */
 473
 474                         goto retry;
 475                 }
 476         }
 477         up_read(&mm->mmap_sem);
 478
 479         mm_rss = get_mm_rss(mm);
 480 #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
 481         mm_rss -= (mm->context.thp_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 482 #endif
 483         if (unlikely(mm_rss >
 484                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 485                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 486 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 487         mm_rss = mm->context.hugetlb_pte_count + mm->context.thp_pte_count;
 488         mm_rss *= REAL_HPAGE_PER_HPAGE;
 489         if (unlikely(mm_rss >
 490                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
 491                 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
 492                         tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 493                 else
 494                         hugetlb_setup(regs);
 495
 496         }
 497 #endif
 498 exit_exception:
 499         exception_exit(prev_state);
 500         return;
 501
 502         /*
 503          * Something tried to access memory that isn't in our memory map..
 504          * Fix it, but check if it's kernel or user first..
 505          */
 506 bad_area:
 507         insn = get_fault_insn(regs, insn);
 508         up_read(&mm->mmap_sem);
 509
 510 handle_kernel_fault:
 511         do_kernel_fault(regs, si_code, fault_code, insn, address);
 512         goto exit_exception;
 513
 514 /*
 515  * We ran out of memory, or some other thing happened to us that made
 516  * us unable to handle the page fault gracefully.
 517  */
 518 out_of_memory:
 519         insn = get_fault_insn(regs, insn);
 520         up_read(&mm->mmap_sem);
 521         if (!(regs->tstate & TSTATE_PRIV)) {
 522                 pagefault_out_of_memory();
 523                 goto exit_exception;
 524         }
 525         goto handle_kernel_fault;
 526
 527 intr_or_no_mm:
 528         insn = get_fault_insn(regs, 0);
 529         goto handle_kernel_fault;
 530
 531 do_sigbus:
 532         insn = get_fault_insn(regs, insn);
 533         up_read(&mm->mmap_sem);
 534
 535         /*
 536          * Send a sigbus, regardless of whether we were in kernel
 537          * or user mode.
 538          */
 539         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
 540
 541         /* Kernel mode? Handle exceptions or die */
 542         if (regs->tstate & TSTATE_PRIV)
 543                 goto handle_kernel_fault;
 544 }