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