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
  25 #include <asm/page.h>
  26 #include <asm/pgtable.h>
  27 #include <asm/openprom.h>
  28 #include <asm/oplib.h>
  29 #include <asm/uaccess.h>
  30 #include <asm/asi.h>
  31 #include <asm/lsu.h>
  32 #include <asm/sections.h>
  33 #include <asm/mmu_context.h>
  34
  35 int show_unhandled_signals = 1;
  36
  37 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
  38 {
  39         int ret = 0;
  40
  41         /* kprobe_running() needs smp_processor_id() */
  42         if (kprobes_built_in() && !user_mode(regs)) {
  43                 preempt_disable();
  44                 if (kprobe_running() && kprobe_fault_handler(regs, 0))
  45                         ret = 1;
  46                 preempt_enable();
  47         }
  48         return ret;
  49 }
  50
  51 static void __kprobes unhandled_fault(unsigned long address,
  52                                       struct task_struct *tsk,
  53                                       struct pt_regs *regs)
  54 {
  55         if ((unsigned long) address < PAGE_SIZE) {
  56                 printk(KERN_ALERT "Unable to handle kernel NULL "
  57                        "pointer dereference\n");
  58         } else {
  59                 printk(KERN_ALERT "Unable to handle kernel paging request "
  60                        "at virtual address %016lx\n", (unsigned long)address);
  61         }
  62         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  63                (tsk->mm ?
  64                 CTX_HWBITS(tsk->mm->context) :
  65                 CTX_HWBITS(tsk->active_mm->context)));
  66         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  67                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  68                           (unsigned long) tsk->active_mm->pgd));
  69         die_if_kernel("Oops", regs);
  70 }
  71
  72 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  73 {
  74         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  75                regs->tpc);
  76         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  77         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  78         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  79         dump_stack();
  80         unhandled_fault(regs->tpc, current, regs);
  81 }
  82
  83 /*
  84  * We now make sure that mmap_sem is held in all paths that call
  85  * this. Additionally, to prevent kswapd from ripping ptes from
  86  * under us, raise interrupts around the time that we look at the
  87  * pte, kswapd will have to wait to get his smp ipi response from
  88  * us. vmtruncate likewise. This saves us having to get pte lock.
  89  */
  90 static unsigned int get_user_insn(unsigned long tpc)
  91 {
  92         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  93         pud_t *pudp;
  94         pmd_t *pmdp;
  95         pte_t *ptep, pte;
  96         unsigned long pa;
  97         u32 insn = 0;
  98         unsigned long pstate;
  99
 100         if (pgd_none(*pgdp))
 101                 goto outret;
 102         pudp = pud_offset(pgdp, tpc);
 103         if (pud_none(*pudp))
 104                 goto outret;
 105         pmdp = pmd_offset(pudp, tpc);
 106         if (pmd_none(*pmdp))
 107                 goto outret;
 108
 109         /* This disables preemption for us as well. */
 110         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 111         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 112                                 : : "r" (pstate), "i" (PSTATE_IE));
 113         ptep = pte_offset_map(pmdp, tpc);
 114         pte = *ptep;
 115         if (!pte_present(pte))
 116                 goto out;
 117
 118         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 119         pa += (tpc & ~PAGE_MASK);
 120
 121         /* Use phys bypass so we don't pollute dtlb/dcache. */
 122         __asm__ __volatile__("lduwa [%1] %2, %0"
 123                              : "=r" (insn)
 124                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 125
 126 out:
 127         pte_unmap(ptep);
 128         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 129 outret:
 130         return insn;
 131 }
 132
 133 static inline void
 134 show_signal_msg(struct pt_regs *regs, int sig, int code,
 135                 unsigned long address, struct task_struct *tsk)
 136 {
 137         if (!unhandled_signal(tsk, sig))
 138                 return;
 139
 140         if (!printk_ratelimit())
 141                 return;
 142
 143         printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
 144                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 145                tsk->comm, task_pid_nr(tsk), address,
 146                (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
 147                (void *)regs->u_regs[UREG_FP], code);
 148
 149         print_vma_addr(KERN_CONT " in ", regs->tpc);
 150
 151         printk(KERN_CONT "\n");
 152 }
 153
 154 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 155
 156 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 157                              unsigned int insn, int fault_code)
 158 {
 159         unsigned long addr;
 160         siginfo_t info;
 161
 162         info.si_code = code;
 163         info.si_signo = sig;
 164         info.si_errno = 0;
 165         if (fault_code & FAULT_CODE_ITLB)
 166                 addr = regs->tpc;
 167         else
 168                 addr = compute_effective_address(regs, insn, 0);
 169         info.si_addr = (void __user *) addr;
 170         info.si_trapno = 0;
 171
 172         if (unlikely(show_unhandled_signals))
 173                 show_signal_msg(regs, sig, code, addr, current);
 174
 175         force_sig_info(sig, &info, current);
 176 }
 177
 178 extern int handle_ldf_stq(u32, struct pt_regs *);
 179 extern int handle_ld_nf(u32, struct pt_regs *);
 180
 181 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 182 {
 183         if (!insn) {
 184                 if (!regs->tpc || (regs->tpc & 0x3))
 185                         return 0;
 186                 if (regs->tstate & TSTATE_PRIV) {
 187                         insn = *(unsigned int *) regs->tpc;
 188                 } else {
 189                         insn = get_user_insn(regs->tpc);
 190                 }
 191         }
 192         return insn;
 193 }
 194
 195 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 196                                       int fault_code, unsigned int insn,
 197                                       unsigned long address)
 198 {
 199         unsigned char asi = ASI_P;
 200
 201         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 202                 goto cannot_handle;
 203
 204         /* If user insn could be read (thus insn is zero), that
 205          * is fine.  We will just gun down the process with a signal
 206          * in that case.
 207          */
 208
 209         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 210             (insn & 0xc0800000) == 0xc0800000) {
 211                 if (insn & 0x2000)
 212                         asi = (regs->tstate >> 24);
 213                 else
 214                         asi = (insn >> 5);
 215                 if ((asi & 0xf2) == 0x82) {
 216                         if (insn & 0x1000000) {
 217                                 handle_ldf_stq(insn, regs);
 218                         } else {
 219                                 /* This was a non-faulting load. Just clear the
 220                                  * destination register(s) and continue with the next
 221                                  * instruction. -jj
 222                                  */
 223                                 handle_ld_nf(insn, regs);
 224                         }
 225                         return;
 226                 }
 227         }
 228
 229         /* Is this in ex_table? */
 230         if (regs->tstate & TSTATE_PRIV) {
 231                 const struct exception_table_entry *entry;
 232
 233                 entry = search_exception_tables(regs->tpc);
 234                 if (entry) {
 235                         regs->tpc = entry->fixup;
 236                         regs->tnpc = regs->tpc + 4;
 237                         return;
 238                 }
 239         } else {
 240                 /* The si_code was set to make clear whether
 241                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 242                  */
 243                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 244                 return;
 245         }
 246
 247 cannot_handle:
 248         unhandled_fault (address, current, regs);
 249 }
 250
 251 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 252 {
 253         static int times;
 254
 255         if (times++ < 10)
 256                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 257                        "64-bit TPC [%lx]\n",
 258                        current->comm, current->pid,
 259                        regs->tpc);
 260         show_regs(regs);
 261 }
 262
 263 static void noinline __kprobes bogus_32bit_fault_address(struct pt_regs *regs,
 264                                                          unsigned long addr)
 265 {
 266         static int times;
 267
 268         if (times++ < 10)
 269                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process "
 270                        "reports 64-bit fault address [%lx]\n",
 271                        current->comm, current->pid, addr);
 272         show_regs(regs);
 273 }
 274
 275 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 276 {
 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
 283         fault_code = get_thread_fault_code();
 284
 285         if (notify_page_fault(regs))
 286                 return;
 287
 288         si_code = SEGV_MAPERR;
 289         address = current_thread_info()->fault_address;
 290
 291         if ((fault_code & FAULT_CODE_ITLB) &&
 292             (fault_code & FAULT_CODE_DTLB))
 293                 BUG();
 294
 295         if (test_thread_flag(TIF_32BIT)) {
 296                 if (!(regs->tstate & TSTATE_PRIV)) {
 297                         if (unlikely((regs->tpc >> 32) != 0)) {
 298                                 bogus_32bit_fault_tpc(regs);
 299                                 goto intr_or_no_mm;
 300                         }
 301                 }
 302                 if (unlikely((address >> 32) != 0)) {
 303                         bogus_32bit_fault_address(regs, address);
 304                         goto intr_or_no_mm;
 305                 }
 306         }
 307
 308         if (regs->tstate & TSTATE_PRIV) {
 309                 unsigned long tpc = regs->tpc;
 310
 311                 /* Sanity check the PC. */
 312                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 313                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 314                         /* Valid, no problems... */
 315                 } else {
 316                         bad_kernel_pc(regs, address);
 317                         return;
 318                 }
 319         }
 320
 321         /*
 322          * If we're in an interrupt or have no user
 323          * context, we must not take the fault..
 324          */
 325         if (in_atomic() || !mm)
 326                 goto intr_or_no_mm;
 327
 328         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 329
 330         if (!down_read_trylock(&mm->mmap_sem)) {
 331                 if ((regs->tstate & TSTATE_PRIV) &&
 332                     !search_exception_tables(regs->tpc)) {
 333                         insn = get_fault_insn(regs, insn);
 334                         goto handle_kernel_fault;
 335                 }
 336                 down_read(&mm->mmap_sem);
 337         }
 338
 339         vma = find_vma(mm, address);
 340         if (!vma)
 341                 goto bad_area;
 342
 343         /* Pure DTLB misses do not tell us whether the fault causing
 344          * load/store/atomic was a write or not, it only says that there
 345          * was no match.  So in such a case we (carefully) read the
 346          * instruction to try and figure this out.  It's an optimization
 347          * so it's ok if we can't do this.
 348          *
 349          * Special hack, window spill/fill knows the exact fault type.
 350          */
 351         if (((fault_code &
 352               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 353             (vma->vm_flags & VM_WRITE) != 0) {
 354                 insn = get_fault_insn(regs, 0);
 355                 if (!insn)
 356                         goto continue_fault;
 357                 /* All loads, stores and atomics have bits 30 and 31 both set
 358                  * in the instruction.  Bit 21 is set in all stores, but we
 359                  * have to avoid prefetches which also have bit 21 set.
 360                  */
 361                 if ((insn & 0xc0200000) == 0xc0200000 &&
 362                     (insn & 0x01780000) != 0x01680000) {
 363                         /* Don't bother updating thread struct value,
 364                          * because update_mmu_cache only cares which tlb
 365                          * the access came from.
 366                          */
 367                         fault_code |= FAULT_CODE_WRITE;
 368                 }
 369         }
 370 continue_fault:
 371
 372         if (vma->vm_start <= address)
 373                 goto good_area;
 374         if (!(vma->vm_flags & VM_GROWSDOWN))
 375                 goto bad_area;
 376         if (!(fault_code & FAULT_CODE_WRITE)) {
 377                 /* Non-faulting loads shouldn't expand stack. */
 378                 insn = get_fault_insn(regs, insn);
 379                 if ((insn & 0xc0800000) == 0xc0800000) {
 380                         unsigned char asi;
 381
 382                         if (insn & 0x2000)
 383                                 asi = (regs->tstate >> 24);
 384                         else
 385                                 asi = (insn >> 5);
 386                         if ((asi & 0xf2) == 0x82)
 387                                 goto bad_area;
 388                 }
 389         }
 390         if (expand_stack(vma, address))
 391                 goto bad_area;
 392         /*
 393          * Ok, we have a good vm_area for this memory access, so
 394          * we can handle it..
 395          */
 396 good_area:
 397         si_code = SEGV_ACCERR;
 398
 399         /* If we took a ITLB miss on a non-executable page, catch
 400          * that here.
 401          */
 402         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 403                 BUG_ON(address != regs->tpc);
 404                 BUG_ON(regs->tstate & TSTATE_PRIV);
 405                 goto bad_area;
 406         }
 407
 408         if (fault_code & FAULT_CODE_WRITE) {
 409                 if (!(vma->vm_flags & VM_WRITE))
 410                         goto bad_area;
 411
 412                 /* Spitfire has an icache which does not snoop
 413                  * processor stores.  Later processors do...
 414                  */
 415                 if (tlb_type == spitfire &&
 416                     (vma->vm_flags & VM_EXEC) != 0 &&
 417                     vma->vm_file != NULL)
 418                         set_thread_fault_code(fault_code |
 419                                               FAULT_CODE_BLKCOMMIT);
 420         } else {
 421                 /* Allow reads even for write-only mappings */
 422                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 423                         goto bad_area;
 424         }
 425
 426         fault = handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE) ? FAULT_FLAG_WRITE : 0);
 427         if (unlikely(fault & VM_FAULT_ERROR)) {
 428                 if (fault & VM_FAULT_OOM)
 429                         goto out_of_memory;
 430                 else if (fault & VM_FAULT_SIGBUS)
 431                         goto do_sigbus;
 432                 BUG();
 433         }
 434         if (fault & VM_FAULT_MAJOR) {
 435                 current->maj_flt++;
 436                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
 437         } else {
 438                 current->min_flt++;
 439                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
 440         }
 441         up_read(&mm->mmap_sem);
 442
 443         mm_rss = get_mm_rss(mm);
 444 #ifdef CONFIG_HUGETLB_PAGE
 445         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 446 #endif
 447         if (unlikely(mm_rss >
 448                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 449                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 450 #ifdef CONFIG_HUGETLB_PAGE
 451         mm_rss = mm->context.huge_pte_count;
 452         if (unlikely(mm_rss >
 453                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
 454                 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 455 #endif
 456         return;
 457
 458         /*
 459          * Something tried to access memory that isn't in our memory map..
 460          * Fix it, but check if it's kernel or user first..
 461          */
 462 bad_area:
 463         insn = get_fault_insn(regs, insn);
 464         up_read(&mm->mmap_sem);
 465
 466 handle_kernel_fault:
 467         do_kernel_fault(regs, si_code, fault_code, insn, address);
 468         return;
 469
 470 /*
 471  * We ran out of memory, or some other thing happened to us that made
 472  * us unable to handle the page fault gracefully.
 473  */
 474 out_of_memory:
 475         insn = get_fault_insn(regs, insn);
 476         up_read(&mm->mmap_sem);
 477         if (!(regs->tstate & TSTATE_PRIV)) {
 478                 pagefault_out_of_memory();
 479                 return;
 480         }
 481         goto handle_kernel_fault;
 482
 483 intr_or_no_mm:
 484         insn = get_fault_insn(regs, 0);
 485         goto handle_kernel_fault;
 486
 487 do_sigbus:
 488         insn = get_fault_insn(regs, insn);
 489         up_read(&mm->mmap_sem);
 490
 491         /*
 492          * Send a sigbus, regardless of whether we were in kernel
 493          * or user mode.
 494          */
 495         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 496
 497         /* Kernel mode? Handle exceptions or die */
 498         if (regs->tstate & TSTATE_PRIV)
 499                 goto handle_kernel_fault;
 500 }