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
  36 int show_unhandled_signals = 1;
  37
  38 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
  39 {
  40         int ret = 0;
  41
  42         /* kprobe_running() needs smp_processor_id() */
  43         if (kprobes_built_in() && !user_mode(regs)) {
  44                 preempt_disable();
  45                 if (kprobe_running() && kprobe_fault_handler(regs, 0))
  46                         ret = 1;
  47                 preempt_enable();
  48         }
  49         return ret;
  50 }
  51
  52 static void __kprobes unhandled_fault(unsigned long address,
  53                                       struct task_struct *tsk,
  54                                       struct pt_regs *regs)
  55 {
  56         if ((unsigned long) address < PAGE_SIZE) {
  57                 printk(KERN_ALERT "Unable to handle kernel NULL "
  58                        "pointer dereference\n");
  59         } else {
  60                 printk(KERN_ALERT "Unable to handle kernel paging request "
  61                        "at virtual address %016lx\n", (unsigned long)address);
  62         }
  63         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  64                (tsk->mm ?
  65                 CTX_HWBITS(tsk->mm->context) :
  66                 CTX_HWBITS(tsk->active_mm->context)));
  67         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  68                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  69                           (unsigned long) tsk->active_mm->pgd));
  70         die_if_kernel("Oops", regs);
  71 }
  72
  73 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  74 {
  75         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  76                regs->tpc);
  77         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  78         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  79         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  80         dump_stack();
  81         unhandled_fault(regs->tpc, current, regs);
  82 }
  83
  84 /*
  85  * We now make sure that mmap_sem is held in all paths that call
  86  * this. Additionally, to prevent kswapd from ripping ptes from
  87  * under us, raise interrupts around the time that we look at the
  88  * pte, kswapd will have to wait to get his smp ipi response from
  89  * us. vmtruncate likewise. This saves us having to get pte lock.
  90  */
  91 static unsigned int get_user_insn(unsigned long tpc)
  92 {
  93         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  94         pud_t *pudp;
  95         pmd_t *pmdp;
  96         pte_t *ptep, pte;
  97         unsigned long pa;
  98         u32 insn = 0;
  99         unsigned long pstate;
 100
 101         if (pgd_none(*pgdp))
 102                 goto outret;
 103         pudp = pud_offset(pgdp, tpc);
 104         if (pud_none(*pudp))
 105                 goto outret;
 106         pmdp = pmd_offset(pudp, tpc);
 107         if (pmd_none(*pmdp))
 108                 goto outret;
 109
 110         /* This disables preemption for us as well. */
 111         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 112         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 113                                 : : "r" (pstate), "i" (PSTATE_IE));
 114         ptep = pte_offset_map(pmdp, tpc);
 115         pte = *ptep;
 116         if (!pte_present(pte))
 117                 goto out;
 118
 119         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 120         pa += (tpc & ~PAGE_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
 127 out:
 128         pte_unmap(ptep);
 129         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 130 outret:
 131         return insn;
 132 }
 133
 134 static inline void
 135 show_signal_msg(struct pt_regs *regs, int sig, int code,
 136                 unsigned long address, struct task_struct *tsk)
 137 {
 138         if (!unhandled_signal(tsk, sig))
 139                 return;
 140
 141         if (!printk_ratelimit())
 142                 return;
 143
 144         printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
 145                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 146                tsk->comm, task_pid_nr(tsk), address,
 147                (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
 148                (void *)regs->u_regs[UREG_FP], code);
 149
 150         print_vma_addr(KERN_CONT " in ", regs->tpc);
 151
 152         printk(KERN_CONT "\n");
 153 }
 154
 155 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 156                              unsigned int insn, int fault_code)
 157 {
 158         unsigned long addr;
 159         siginfo_t info;
 160
 161         info.si_code = code;
 162         info.si_signo = sig;
 163         info.si_errno = 0;
 164         if (fault_code & FAULT_CODE_ITLB)
 165                 addr = regs->tpc;
 166         else
 167                 addr = compute_effective_address(regs, insn, 0);
 168         info.si_addr = (void __user *) addr;
 169         info.si_trapno = 0;
 170
 171         if (unlikely(show_unhandled_signals))
 172                 show_signal_msg(regs, sig, code, addr, current);
 173
 174         force_sig_info(sig, &info, current);
 175 }
 176
 177 extern int handle_ldf_stq(u32, struct pt_regs *);
 178 extern int handle_ld_nf(u32, struct pt_regs *);
 179
 180 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 181 {
 182         if (!insn) {
 183                 if (!regs->tpc || (regs->tpc & 0x3))
 184                         return 0;
 185                 if (regs->tstate & TSTATE_PRIV) {
 186                         insn = *(unsigned int *) regs->tpc;
 187                 } else {
 188                         insn = get_user_insn(regs->tpc);
 189                 }
 190         }
 191         return insn;
 192 }
 193
 194 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 195                                       int fault_code, unsigned int insn,
 196                                       unsigned long address)
 197 {
 198         unsigned char asi = ASI_P;
 199
 200         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 201                 goto cannot_handle;
 202
 203         /* If user insn could be read (thus insn is zero), that
 204          * is fine.  We will just gun down the process with a signal
 205          * in that case.
 206          */
 207
 208         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 209             (insn & 0xc0800000) == 0xc0800000) {
 210                 if (insn & 0x2000)
 211                         asi = (regs->tstate >> 24);
 212                 else
 213                         asi = (insn >> 5);
 214                 if ((asi & 0xf2) == 0x82) {
 215                         if (insn & 0x1000000) {
 216                                 handle_ldf_stq(insn, regs);
 217                         } else {
 218                                 /* This was a non-faulting load. Just clear the
 219                                  * destination register(s) and continue with the next
 220                                  * instruction. -jj
 221                                  */
 222                                 handle_ld_nf(insn, regs);
 223                         }
 224                         return;
 225                 }
 226         }
 227
 228         /* Is this in ex_table? */
 229         if (regs->tstate & TSTATE_PRIV) {
 230                 const struct exception_table_entry *entry;
 231
 232                 entry = search_exception_tables(regs->tpc);
 233                 if (entry) {
 234                         regs->tpc = entry->fixup;
 235                         regs->tnpc = regs->tpc + 4;
 236                         return;
 237                 }
 238         } else {
 239                 /* The si_code was set to make clear whether
 240                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 241                  */
 242                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 243                 return;
 244         }
 245
 246 cannot_handle:
 247         unhandled_fault (address, current, regs);
 248 }
 249
 250 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 251 {
 252         static int times;
 253
 254         if (times++ < 10)
 255                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 256                        "64-bit TPC [%lx]\n",
 257                        current->comm, current->pid,
 258                        regs->tpc);
 259         show_regs(regs);
 260 }
 261
 262 static void noinline __kprobes bogus_32bit_fault_address(struct pt_regs *regs,
 263                                                          unsigned long addr)
 264 {
 265         static int times;
 266
 267         if (times++ < 10)
 268                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process "
 269                        "reports 64-bit fault address [%lx]\n",
 270                        current->comm, current->pid, addr);
 271         show_regs(regs);
 272 }
 273
 274 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 275 {
 276         enum ctx_state prev_state = exception_enter();
 277         struct mm_struct *mm = current->mm;
 278         struct vm_area_struct *vma;
 279         unsigned int insn = 0;
 280         int si_code, fault_code, fault;
 281         unsigned long address, mm_rss;
 282         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 283
 284         fault_code = get_thread_fault_code();
 285
 286         if (notify_page_fault(regs))
 287                 goto exit_exception;
 288
 289         si_code = SEGV_MAPERR;
 290         address = current_thread_info()->fault_address;
 291
 292         if ((fault_code & FAULT_CODE_ITLB) &&
 293             (fault_code & FAULT_CODE_DTLB))
 294                 BUG();
 295
 296         if (test_thread_flag(TIF_32BIT)) {
 297                 if (!(regs->tstate & TSTATE_PRIV)) {
 298                         if (unlikely((regs->tpc >> 32) != 0)) {
 299                                 bogus_32bit_fault_tpc(regs);
 300                                 goto intr_or_no_mm;
 301                         }
 302                 }
 303                 if (unlikely((address >> 32) != 0)) {
 304                         bogus_32bit_fault_address(regs, address);
 305                         goto intr_or_no_mm;
 306                 }
 307         }
 308
 309         if (regs->tstate & TSTATE_PRIV) {
 310                 unsigned long tpc = regs->tpc;
 311
 312                 /* Sanity check the PC. */
 313                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 314                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 315                         /* Valid, no problems... */
 316                 } else {
 317                         bad_kernel_pc(regs, address);
 318                         goto exit_exception;
 319                 }
 320         } else
 321                 flags |= FAULT_FLAG_USER;
 322
 323         /*
 324          * If we're in an interrupt or have no user
 325          * context, we must not take the fault..
 326          */
 327         if (in_atomic() || !mm)
 328                 goto intr_or_no_mm;
 329
 330         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 331
 332         if (!down_read_trylock(&mm->mmap_sem)) {
 333                 if ((regs->tstate & TSTATE_PRIV) &&
 334                     !search_exception_tables(regs->tpc)) {
 335                         insn = get_fault_insn(regs, insn);
 336                         goto handle_kernel_fault;
 337                 }
 338
 339 retry:
 340                 down_read(&mm->mmap_sem);
 341         }
 342
 343         vma = find_vma(mm, address);
 344         if (!vma)
 345                 goto bad_area;
 346
 347         /* Pure DTLB misses do not tell us whether the fault causing
 348          * load/store/atomic was a write or not, it only says that there
 349          * was no match.  So in such a case we (carefully) read the
 350          * instruction to try and figure this out.  It's an optimization
 351          * so it's ok if we can't do this.
 352          *
 353          * Special hack, window spill/fill knows the exact fault type.
 354          */
 355         if (((fault_code &
 356               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 357             (vma->vm_flags & VM_WRITE) != 0) {
 358                 insn = get_fault_insn(regs, 0);
 359                 if (!insn)
 360                         goto continue_fault;
 361                 /* All loads, stores and atomics have bits 30 and 31 both set
 362                  * in the instruction.  Bit 21 is set in all stores, but we
 363                  * have to avoid prefetches which also have bit 21 set.
 364                  */
 365                 if ((insn & 0xc0200000) == 0xc0200000 &&
 366                     (insn & 0x01780000) != 0x01680000) {
 367                         /* Don't bother updating thread struct value,
 368                          * because update_mmu_cache only cares which tlb
 369                          * the access came from.
 370                          */
 371                         fault_code |= FAULT_CODE_WRITE;
 372                 }
 373         }
 374 continue_fault:
 375
 376         if (vma->vm_start <= address)
 377                 goto good_area;
 378         if (!(vma->vm_flags & VM_GROWSDOWN))
 379                 goto bad_area;
 380         if (!(fault_code & FAULT_CODE_WRITE)) {
 381                 /* Non-faulting loads shouldn't expand stack. */
 382                 insn = get_fault_insn(regs, insn);
 383                 if ((insn & 0xc0800000) == 0xc0800000) {
 384                         unsigned char asi;
 385
 386                         if (insn & 0x2000)
 387                                 asi = (regs->tstate >> 24);
 388                         else
 389                                 asi = (insn >> 5);
 390                         if ((asi & 0xf2) == 0x82)
 391                                 goto bad_area;
 392                 }
 393         }
 394         if (expand_stack(vma, address))
 395                 goto bad_area;
 396         /*
 397          * Ok, we have a good vm_area for this memory access, so
 398          * we can handle it..
 399          */
 400 good_area:
 401         si_code = SEGV_ACCERR;
 402
 403         /* If we took a ITLB miss on a non-executable page, catch
 404          * that here.
 405          */
 406         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 407                 BUG_ON(address != regs->tpc);
 408                 BUG_ON(regs->tstate & TSTATE_PRIV);
 409                 goto bad_area;
 410         }
 411
 412         if (fault_code & FAULT_CODE_WRITE) {
 413                 if (!(vma->vm_flags & VM_WRITE))
 414                         goto bad_area;
 415
 416                 /* Spitfire has an icache which does not snoop
 417                  * processor stores.  Later processors do...
 418                  */
 419                 if (tlb_type == spitfire &&
 420                     (vma->vm_flags & VM_EXEC) != 0 &&
 421                     vma->vm_file != NULL)
 422                         set_thread_fault_code(fault_code |
 423                                               FAULT_CODE_BLKCOMMIT);
 424
 425                 flags |= FAULT_FLAG_WRITE;
 426         } else {
 427                 /* Allow reads even for write-only mappings */
 428                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 429                         goto bad_area;
 430         }
 431
 432         fault = handle_mm_fault(mm, vma, address, flags);
 433
 434         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 435                 goto exit_exception;
 436
 437         if (unlikely(fault & VM_FAULT_ERROR)) {
 438                 if (fault & VM_FAULT_OOM)
 439                         goto out_of_memory;
 440                 else if (fault & VM_FAULT_SIGBUS)
 441                         goto do_sigbus;
 442                 BUG();
 443         }
 444
 445         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 446                 if (fault & VM_FAULT_MAJOR) {
 447                         current->maj_flt++;
 448                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
 449                                       1, regs, address);
 450                 } else {
 451                         current->min_flt++;
 452                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
 453                                       1, regs, address);
 454                 }
 455                 if (fault & VM_FAULT_RETRY) {
 456                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 457                         flags |= FAULT_FLAG_TRIED;
 458
 459                         /* No need to up_read(&mm->mmap_sem) as we would
 460                          * have already released it in __lock_page_or_retry
 461                          * in mm/filemap.c.
 462                          */
 463
 464                         goto retry;
 465                 }
 466         }
 467         up_read(&mm->mmap_sem);
 468
 469         mm_rss = get_mm_rss(mm);
 470 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 471         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 472 #endif
 473         if (unlikely(mm_rss >
 474                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 475                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 476 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 477         mm_rss = mm->context.huge_pte_count;
 478         if (unlikely(mm_rss >
 479                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
 480                 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
 481                         tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 482                 else
 483                         hugetlb_setup(regs);
 484
 485         }
 486 #endif
 487 exit_exception:
 488         exception_exit(prev_state);
 489         return;
 490
 491         /*
 492          * Something tried to access memory that isn't in our memory map..
 493          * Fix it, but check if it's kernel or user first..
 494          */
 495 bad_area:
 496         insn = get_fault_insn(regs, insn);
 497         up_read(&mm->mmap_sem);
 498
 499 handle_kernel_fault:
 500         do_kernel_fault(regs, si_code, fault_code, insn, address);
 501         goto exit_exception;
 502
 503 /*
 504  * We ran out of memory, or some other thing happened to us that made
 505  * us unable to handle the page fault gracefully.
 506  */
 507 out_of_memory:
 508         insn = get_fault_insn(regs, insn);
 509         up_read(&mm->mmap_sem);
 510         if (!(regs->tstate & TSTATE_PRIV)) {
 511                 pagefault_out_of_memory();
 512                 goto exit_exception;
 513         }
 514         goto handle_kernel_fault;
 515
 516 intr_or_no_mm:
 517         insn = get_fault_insn(regs, 0);
 518         goto handle_kernel_fault;
 519
 520 do_sigbus:
 521         insn = get_fault_insn(regs, insn);
 522         up_read(&mm->mmap_sem);
 523
 524         /*
 525          * Send a sigbus, regardless of whether we were in kernel
 526          * or user mode.
 527          */
 528         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 529
 530         /* Kernel mode? Handle exceptions or die */
 531         if (regs->tstate & TSTATE_PRIV)
 532                 goto handle_kernel_fault;
 533 }