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
  26 #include <asm/page.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/openprom.h>
  29 #include <asm/oplib.h>
  30 #include <asm/uaccess.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 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 115         if (pmd_trans_huge(*pmdp)) {
 116                 if (pmd_trans_splitting(*pmdp))
 117                         goto out_irq_enable;
 118
 119                 pa  = pmd_pfn(*pmdp) << PAGE_SHIFT;
 120                 pa += tpc & ~HPAGE_MASK;
 121
 122                 /* Use phys bypass so we don't pollute dtlb/dcache. */
 123                 __asm__ __volatile__("lduwa [%1] %2, %0"
 124                                      : "=r" (insn)
 125                                      : "r" (pa), "i" (ASI_PHYS_USE_EC));
 126         } else
 127 #endif
 128         {
 129                 ptep = pte_offset_map(pmdp, tpc);
 130                 pte = *ptep;
 131                 if (pte_present(pte)) {
 132                         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 133                         pa += (tpc & ~PAGE_MASK);
 134
 135                         /* Use phys bypass so we don't pollute dtlb/dcache. */
 136                         __asm__ __volatile__("lduwa [%1] %2, %0"
 137                                              : "=r" (insn)
 138                                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 139                 }
 140                 pte_unmap(ptep);
 141         }
 142 out_irq_enable:
 143         local_irq_enable();
 144 out:
 145         return insn;
 146 }
 147
 148 static inline void
 149 show_signal_msg(struct pt_regs *regs, int sig, int code,
 150                 unsigned long address, struct task_struct *tsk)
 151 {
 152         if (!unhandled_signal(tsk, sig))
 153                 return;
 154
 155         if (!printk_ratelimit())
 156                 return;
 157
 158         printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
 159                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 160                tsk->comm, task_pid_nr(tsk), address,
 161                (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
 162                (void *)regs->u_regs[UREG_FP], code);
 163
 164         print_vma_addr(KERN_CONT " in ", regs->tpc);
 165
 166         printk(KERN_CONT "\n");
 167 }
 168
 169 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 170                              unsigned long fault_addr, unsigned int insn,
 171                              int fault_code)
 172 {
 173         unsigned long addr;
 174         siginfo_t info;
 175
 176         info.si_code = code;
 177         info.si_signo = sig;
 178         info.si_errno = 0;
 179         if (fault_code & FAULT_CODE_ITLB) {
 180                 addr = regs->tpc;
 181         } else {
 182                 /* If we were able to probe the faulting instruction, use it
 183                  * to compute a precise fault address.  Otherwise use the fault
 184                  * time provided address which may only have page granularity.
 185                  */
 186                 if (insn)
 187                         addr = compute_effective_address(regs, insn, 0);
 188                 else
 189                         addr = fault_addr;
 190         }
 191         info.si_addr = (void __user *) addr;
 192         info.si_trapno = 0;
 193
 194         if (unlikely(show_unhandled_signals))
 195                 show_signal_msg(regs, sig, code, addr, current);
 196
 197         force_sig_info(sig, &info, current);
 198 }
 199
 200 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 201 {
 202         if (!insn) {
 203                 if (!regs->tpc || (regs->tpc & 0x3))
 204                         return 0;
 205                 if (regs->tstate & TSTATE_PRIV) {
 206                         insn = *(unsigned int *) regs->tpc;
 207                 } else {
 208                         insn = get_user_insn(regs->tpc);
 209                 }
 210         }
 211         return insn;
 212 }
 213
 214 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 215                                       int fault_code, unsigned int insn,
 216                                       unsigned long address)
 217 {
 218         unsigned char asi = ASI_P;
 219
 220         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 221                 goto cannot_handle;
 222
 223         /* If user insn could be read (thus insn is zero), that
 224          * is fine.  We will just gun down the process with a signal
 225          * in that case.
 226          */
 227
 228         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 229             (insn & 0xc0800000) == 0xc0800000) {
 230                 if (insn & 0x2000)
 231                         asi = (regs->tstate >> 24);
 232                 else
 233                         asi = (insn >> 5);
 234                 if ((asi & 0xf2) == 0x82) {
 235                         if (insn & 0x1000000) {
 236                                 handle_ldf_stq(insn, regs);
 237                         } else {
 238                                 /* This was a non-faulting load. Just clear the
 239                                  * destination register(s) and continue with the next
 240                                  * instruction. -jj
 241                                  */
 242                                 handle_ld_nf(insn, regs);
 243                         }
 244                         return;
 245                 }
 246         }
 247
 248         /* Is this in ex_table? */
 249         if (regs->tstate & TSTATE_PRIV) {
 250                 const struct exception_table_entry *entry;
 251
 252                 entry = search_exception_tables(regs->tpc);
 253                 if (entry) {
 254                         regs->tpc = entry->fixup;
 255                         regs->tnpc = regs->tpc + 4;
 256                         return;
 257                 }
 258         } else {
 259                 /* The si_code was set to make clear whether
 260                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 261                  */
 262                 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
 263                 return;
 264         }
 265
 266 cannot_handle:
 267         unhandled_fault (address, current, regs);
 268 }
 269
 270 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 271 {
 272         static int times;
 273
 274         if (times++ < 10)
 275                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 276                        "64-bit TPC [%lx]\n",
 277                        current->comm, current->pid,
 278                        regs->tpc);
 279         show_regs(regs);
 280 }
 281
 282 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 283 {
 284         enum ctx_state prev_state = exception_enter();
 285         struct mm_struct *mm = current->mm;
 286         struct vm_area_struct *vma;
 287         unsigned int insn = 0;
 288         int si_code, fault_code, fault;
 289         unsigned long address, mm_rss;
 290         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 291
 292         fault_code = get_thread_fault_code();
 293
 294         if (notify_page_fault(regs))
 295                 goto exit_exception;
 296
 297         si_code = SEGV_MAPERR;
 298         address = current_thread_info()->fault_address;
 299
 300         if ((fault_code & FAULT_CODE_ITLB) &&
 301             (fault_code & FAULT_CODE_DTLB))
 302                 BUG();
 303
 304         if (test_thread_flag(TIF_32BIT)) {
 305                 if (!(regs->tstate & TSTATE_PRIV)) {
 306                         if (unlikely((regs->tpc >> 32) != 0)) {
 307                                 bogus_32bit_fault_tpc(regs);
 308                                 goto intr_or_no_mm;
 309                         }
 310                 }
 311                 if (unlikely((address >> 32) != 0))
 312                         goto intr_or_no_mm;
 313         }
 314
 315         if (regs->tstate & TSTATE_PRIV) {
 316                 unsigned long tpc = regs->tpc;
 317
 318                 /* Sanity check the PC. */
 319                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 320                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 321                         /* Valid, no problems... */
 322                 } else {
 323                         bad_kernel_pc(regs, address);
 324                         goto exit_exception;
 325                 }
 326         } else
 327                 flags |= FAULT_FLAG_USER;
 328
 329         /*
 330          * If we're in an interrupt or have no user
 331          * context, we must not take the fault..
 332          */
 333         if (in_atomic() || !mm)
 334                 goto intr_or_no_mm;
 335
 336         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 337
 338         if (!down_read_trylock(&mm->mmap_sem)) {
 339                 if ((regs->tstate & TSTATE_PRIV) &&
 340                     !search_exception_tables(regs->tpc)) {
 341                         insn = get_fault_insn(regs, insn);
 342                         goto handle_kernel_fault;
 343                 }
 344
 345 retry:
 346                 down_read(&mm->mmap_sem);
 347         }
 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                 BUG_ON(address != regs->tpc);
 414                 BUG_ON(regs->tstate & TSTATE_PRIV);
 415                 goto bad_area;
 416         }
 417
 418         if (fault_code & FAULT_CODE_WRITE) {
 419                 if (!(vma->vm_flags & VM_WRITE))
 420                         goto bad_area;
 421
 422                 /* Spitfire has an icache which does not snoop
 423                  * processor stores.  Later processors do...
 424                  */
 425                 if (tlb_type == spitfire &&
 426                     (vma->vm_flags & VM_EXEC) != 0 &&
 427                     vma->vm_file != NULL)
 428                         set_thread_fault_code(fault_code |
 429                                               FAULT_CODE_BLKCOMMIT);
 430
 431                 flags |= FAULT_FLAG_WRITE;
 432         } else {
 433                 /* Allow reads even for write-only mappings */
 434                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 435                         goto bad_area;
 436         }
 437
 438         fault = handle_mm_fault(mm, vma, address, flags);
 439
 440         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 441                 goto exit_exception;
 442
 443         if (unlikely(fault & VM_FAULT_ERROR)) {
 444                 if (fault & VM_FAULT_OOM)
 445                         goto out_of_memory;
 446                 else if (fault & VM_FAULT_SIGBUS)
 447                         goto do_sigbus;
 448                 BUG();
 449         }
 450
 451         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 452                 if (fault & VM_FAULT_MAJOR) {
 453                         current->maj_flt++;
 454                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
 455                                       1, regs, address);
 456                 } else {
 457                         current->min_flt++;
 458                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
 459                                       1, regs, address);
 460                 }
 461                 if (fault & VM_FAULT_RETRY) {
 462                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 463                         flags |= FAULT_FLAG_TRIED;
 464
 465                         /* No need to up_read(&mm->mmap_sem) as we would
 466                          * have already released it in __lock_page_or_retry
 467                          * in mm/filemap.c.
 468                          */
 469
 470                         goto retry;
 471                 }
 472         }
 473         up_read(&mm->mmap_sem);
 474
 475         mm_rss = get_mm_rss(mm);
 476 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 477         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 478 #endif
 479         if (unlikely(mm_rss >
 480                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 481                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 482 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 483         mm_rss = mm->context.huge_pte_count;
 484         if (unlikely(mm_rss >
 485                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
 486                 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
 487                         tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 488                 else
 489                         hugetlb_setup(regs);
 490
 491         }
 492 #endif
 493 exit_exception:
 494         exception_exit(prev_state);
 495         return;
 496
 497         /*
 498          * Something tried to access memory that isn't in our memory map..
 499          * Fix it, but check if it's kernel or user first..
 500          */
 501 bad_area:
 502         insn = get_fault_insn(regs, insn);
 503         up_read(&mm->mmap_sem);
 504
 505 handle_kernel_fault:
 506         do_kernel_fault(regs, si_code, fault_code, insn, address);
 507         goto exit_exception;
 508
 509 /*
 510  * We ran out of memory, or some other thing happened to us that made
 511  * us unable to handle the page fault gracefully.
 512  */
 513 out_of_memory:
 514         insn = get_fault_insn(regs, insn);
 515         up_read(&mm->mmap_sem);
 516         if (!(regs->tstate & TSTATE_PRIV)) {
 517                 pagefault_out_of_memory();
 518                 goto exit_exception;
 519         }
 520         goto handle_kernel_fault;
 521
 522 intr_or_no_mm:
 523         insn = get_fault_insn(regs, 0);
 524         goto handle_kernel_fault;
 525
 526 do_sigbus:
 527         insn = get_fault_insn(regs, insn);
 528         up_read(&mm->mmap_sem);
 529
 530         /*
 531          * Send a sigbus, regardless of whether we were in kernel
 532          * or user mode.
 533          */
 534         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
 535
 536         /* Kernel mode? Handle exceptions or die */
 537         if (regs->tstate & TSTATE_PRIV)
 538                 goto handle_kernel_fault;
 539 }