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 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 155                              unsigned int insn, int fault_code)
 156 {
 157         unsigned long addr;
 158         siginfo_t info;
 159
 160         info.si_code = code;
 161         info.si_signo = sig;
 162         info.si_errno = 0;
 163         if (fault_code & FAULT_CODE_ITLB)
 164                 addr = regs->tpc;
 165         else
 166                 addr = compute_effective_address(regs, insn, 0);
 167         info.si_addr = (void __user *) addr;
 168         info.si_trapno = 0;
 169
 170         if (unlikely(show_unhandled_signals))
 171                 show_signal_msg(regs, sig, code, addr, current);
 172
 173         force_sig_info(sig, &info, current);
 174 }
 175
 176 extern int handle_ldf_stq(u32, struct pt_regs *);
 177 extern int handle_ld_nf(u32, struct pt_regs *);
 178
 179 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 180 {
 181         if (!insn) {
 182                 if (!regs->tpc || (regs->tpc & 0x3))
 183                         return 0;
 184                 if (regs->tstate & TSTATE_PRIV) {
 185                         insn = *(unsigned int *) regs->tpc;
 186                 } else {
 187                         insn = get_user_insn(regs->tpc);
 188                 }
 189         }
 190         return insn;
 191 }
 192
 193 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 194                                       int fault_code, unsigned int insn,
 195                                       unsigned long address)
 196 {
 197         unsigned char asi = ASI_P;
 198
 199         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 200                 goto cannot_handle;
 201
 202         /* If user insn could be read (thus insn is zero), that
 203          * is fine.  We will just gun down the process with a signal
 204          * in that case.
 205          */
 206
 207         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 208             (insn & 0xc0800000) == 0xc0800000) {
 209                 if (insn & 0x2000)
 210                         asi = (regs->tstate >> 24);
 211                 else
 212                         asi = (insn >> 5);
 213                 if ((asi & 0xf2) == 0x82) {
 214                         if (insn & 0x1000000) {
 215                                 handle_ldf_stq(insn, regs);
 216                         } else {
 217                                 /* This was a non-faulting load. Just clear the
 218                                  * destination register(s) and continue with the next
 219                                  * instruction. -jj
 220                                  */
 221                                 handle_ld_nf(insn, regs);
 222                         }
 223                         return;
 224                 }
 225         }
 226
 227         /* Is this in ex_table? */
 228         if (regs->tstate & TSTATE_PRIV) {
 229                 const struct exception_table_entry *entry;
 230
 231                 entry = search_exception_tables(regs->tpc);
 232                 if (entry) {
 233                         regs->tpc = entry->fixup;
 234                         regs->tnpc = regs->tpc + 4;
 235                         return;
 236                 }
 237         } else {
 238                 /* The si_code was set to make clear whether
 239                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 240                  */
 241                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 242                 return;
 243         }
 244
 245 cannot_handle:
 246         unhandled_fault (address, current, regs);
 247 }
 248
 249 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 250 {
 251         static int times;
 252
 253         if (times++ < 10)
 254                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 255                        "64-bit TPC [%lx]\n",
 256                        current->comm, current->pid,
 257                        regs->tpc);
 258         show_regs(regs);
 259 }
 260
 261 static void noinline __kprobes bogus_32bit_fault_address(struct pt_regs *regs,
 262                                                          unsigned long addr)
 263 {
 264         static int times;
 265
 266         if (times++ < 10)
 267                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process "
 268                        "reports 64-bit fault address [%lx]\n",
 269                        current->comm, current->pid, addr);
 270         show_regs(regs);
 271 }
 272
 273 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 274 {
 275         struct mm_struct *mm = current->mm;
 276         struct vm_area_struct *vma;
 277         unsigned int insn = 0;
 278         int si_code, fault_code, fault;
 279         unsigned long address, mm_rss;
 280         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 281
 282         fault_code = get_thread_fault_code();
 283
 284         if (notify_page_fault(regs))
 285                 return;
 286
 287         si_code = SEGV_MAPERR;
 288         address = current_thread_info()->fault_address;
 289
 290         if ((fault_code & FAULT_CODE_ITLB) &&
 291             (fault_code & FAULT_CODE_DTLB))
 292                 BUG();
 293
 294         if (test_thread_flag(TIF_32BIT)) {
 295                 if (!(regs->tstate & TSTATE_PRIV)) {
 296                         if (unlikely((regs->tpc >> 32) != 0)) {
 297                                 bogus_32bit_fault_tpc(regs);
 298                                 goto intr_or_no_mm;
 299                         }
 300                 }
 301                 if (unlikely((address >> 32) != 0)) {
 302                         bogus_32bit_fault_address(regs, address);
 303                         goto intr_or_no_mm;
 304                 }
 305         }
 306
 307         if (regs->tstate & TSTATE_PRIV) {
 308                 unsigned long tpc = regs->tpc;
 309
 310                 /* Sanity check the PC. */
 311                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 312                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 313                         /* Valid, no problems... */
 314                 } else {
 315                         bad_kernel_pc(regs, address);
 316                         return;
 317                 }
 318         }
 319
 320         /*
 321          * If we're in an interrupt or have no user
 322          * context, we must not take the fault..
 323          */
 324         if (in_atomic() || !mm)
 325                 goto intr_or_no_mm;
 326
 327         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 328
 329         if (!down_read_trylock(&mm->mmap_sem)) {
 330                 if ((regs->tstate & TSTATE_PRIV) &&
 331                     !search_exception_tables(regs->tpc)) {
 332                         insn = get_fault_insn(regs, insn);
 333                         goto handle_kernel_fault;
 334                 }
 335
 336 retry:
 337                 down_read(&mm->mmap_sem);
 338         }
 339
 340         vma = find_vma(mm, address);
 341         if (!vma)
 342                 goto bad_area;
 343
 344         /* Pure DTLB misses do not tell us whether the fault causing
 345          * load/store/atomic was a write or not, it only says that there
 346          * was no match.  So in such a case we (carefully) read the
 347          * instruction to try and figure this out.  It's an optimization
 348          * so it's ok if we can't do this.
 349          *
 350          * Special hack, window spill/fill knows the exact fault type.
 351          */
 352         if (((fault_code &
 353               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 354             (vma->vm_flags & VM_WRITE) != 0) {
 355                 insn = get_fault_insn(regs, 0);
 356                 if (!insn)
 357                         goto continue_fault;
 358                 /* All loads, stores and atomics have bits 30 and 31 both set
 359                  * in the instruction.  Bit 21 is set in all stores, but we
 360                  * have to avoid prefetches which also have bit 21 set.
 361                  */
 362                 if ((insn & 0xc0200000) == 0xc0200000 &&
 363                     (insn & 0x01780000) != 0x01680000) {
 364                         /* Don't bother updating thread struct value,
 365                          * because update_mmu_cache only cares which tlb
 366                          * the access came from.
 367                          */
 368                         fault_code |= FAULT_CODE_WRITE;
 369                 }
 370         }
 371 continue_fault:
 372
 373         if (vma->vm_start <= address)
 374                 goto good_area;
 375         if (!(vma->vm_flags & VM_GROWSDOWN))
 376                 goto bad_area;
 377         if (!(fault_code & FAULT_CODE_WRITE)) {
 378                 /* Non-faulting loads shouldn't expand stack. */
 379                 insn = get_fault_insn(regs, insn);
 380                 if ((insn & 0xc0800000) == 0xc0800000) {
 381                         unsigned char asi;
 382
 383                         if (insn & 0x2000)
 384                                 asi = (regs->tstate >> 24);
 385                         else
 386                                 asi = (insn >> 5);
 387                         if ((asi & 0xf2) == 0x82)
 388                                 goto bad_area;
 389                 }
 390         }
 391         if (expand_stack(vma, address))
 392                 goto bad_area;
 393         /*
 394          * Ok, we have a good vm_area for this memory access, so
 395          * we can handle it..
 396          */
 397 good_area:
 398         si_code = SEGV_ACCERR;
 399
 400         /* If we took a ITLB miss on a non-executable page, catch
 401          * that here.
 402          */
 403         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 404                 BUG_ON(address != regs->tpc);
 405                 BUG_ON(regs->tstate & TSTATE_PRIV);
 406                 goto bad_area;
 407         }
 408
 409         if (fault_code & FAULT_CODE_WRITE) {
 410                 if (!(vma->vm_flags & VM_WRITE))
 411                         goto bad_area;
 412
 413                 /* Spitfire has an icache which does not snoop
 414                  * processor stores.  Later processors do...
 415                  */
 416                 if (tlb_type == spitfire &&
 417                     (vma->vm_flags & VM_EXEC) != 0 &&
 418                     vma->vm_file != NULL)
 419                         set_thread_fault_code(fault_code |
 420                                               FAULT_CODE_BLKCOMMIT);
 421         } else {
 422                 /* Allow reads even for write-only mappings */
 423                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 424                         goto bad_area;
 425         }
 426
 427         flags |= ((fault_code & FAULT_CODE_WRITE) ? FAULT_FLAG_WRITE : 0);
 428         fault = handle_mm_fault(mm, vma, address, flags);
 429
 430         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 431                 return;
 432
 433         if (unlikely(fault & VM_FAULT_ERROR)) {
 434                 if (fault & VM_FAULT_OOM)
 435                         goto out_of_memory;
 436                 else if (fault & VM_FAULT_SIGBUS)
 437                         goto do_sigbus;
 438                 BUG();
 439         }
 440
 441         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 442                 if (fault & VM_FAULT_MAJOR) {
 443                         current->maj_flt++;
 444                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
 445                                       1, regs, address);
 446                 } else {
 447                         current->min_flt++;
 448                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
 449                                       1, regs, address);
 450                 }
 451                 if (fault & VM_FAULT_RETRY) {
 452                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 453                         flags |= FAULT_FLAG_TRIED;
 454
 455                         /* No need to up_read(&mm->mmap_sem) as we would
 456                          * have already released it in __lock_page_or_retry
 457                          * in mm/filemap.c.
 458                          */
 459
 460                         goto retry;
 461                 }
 462         }
 463         up_read(&mm->mmap_sem);
 464
 465         mm_rss = get_mm_rss(mm);
 466 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 467         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 468 #endif
 469         if (unlikely(mm_rss >
 470                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 471                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 472 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 473         mm_rss = mm->context.huge_pte_count;
 474         if (unlikely(mm_rss >
 475                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
 476                 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
 477                         tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 478                 else
 479                         hugetlb_setup(regs);
 480
 481         }
 482 #endif
 483         return;
 484
 485         /*
 486          * Something tried to access memory that isn't in our memory map..
 487          * Fix it, but check if it's kernel or user first..
 488          */
 489 bad_area:
 490         insn = get_fault_insn(regs, insn);
 491         up_read(&mm->mmap_sem);
 492
 493 handle_kernel_fault:
 494         do_kernel_fault(regs, si_code, fault_code, insn, address);
 495         return;
 496
 497 /*
 498  * We ran out of memory, or some other thing happened to us that made
 499  * us unable to handle the page fault gracefully.
 500  */
 501 out_of_memory:
 502         insn = get_fault_insn(regs, insn);
 503         up_read(&mm->mmap_sem);
 504         if (!(regs->tstate & TSTATE_PRIV)) {
 505                 pagefault_out_of_memory();
 506                 return;
 507         }
 508         goto handle_kernel_fault;
 509
 510 intr_or_no_mm:
 511         insn = get_fault_insn(regs, 0);
 512         goto handle_kernel_fault;
 513
 514 do_sigbus:
 515         insn = get_fault_insn(regs, insn);
 516         up_read(&mm->mmap_sem);
 517
 518         /*
 519          * Send a sigbus, regardless of whether we were in kernel
 520          * or user mode.
 521          */
 522         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 523
 524         /* Kernel mode? Handle exceptions or die */
 525         if (regs->tstate & TSTATE_PRIV)
 526                 goto handle_kernel_fault;
 527 }