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/interrupt.h>
  20 #include <linux/kprobes.h>
  21 #include <linux/kdebug.h>
  22 #include <linux/percpu.h>
  23
  24 #include <asm/page.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/openprom.h>
  27 #include <asm/oplib.h>
  28 #include <asm/uaccess.h>
  29 #include <asm/asi.h>
  30 #include <asm/lsu.h>
  31 #include <asm/sections.h>
  32 #include <asm/mmu_context.h>
  33
  34 #ifdef CONFIG_KPROBES
  35 static inline int notify_page_fault(struct pt_regs *regs)
  36 {
  37         int ret = 0;
  38
  39         /* kprobe_running() needs smp_processor_id() */
  40         if (!user_mode(regs)) {
  41                 preempt_disable();
  42                 if (kprobe_running() && kprobe_fault_handler(regs, 0))
  43                         ret = 1;
  44                 preempt_enable();
  45         }
  46         return ret;
  47 }
  48 #else
  49 static inline int notify_page_fault(struct pt_regs *regs)
  50 {
  51         return 0;
  52 }
  53 #endif
  54
  55 static void __kprobes unhandled_fault(unsigned long address,
  56                                       struct task_struct *tsk,
  57                                       struct pt_regs *regs)
  58 {
  59         if ((unsigned long) address < PAGE_SIZE) {
  60                 printk(KERN_ALERT "Unable to handle kernel NULL "
  61                        "pointer dereference\n");
  62         } else {
  63                 printk(KERN_ALERT "Unable to handle kernel paging request "
  64                        "at virtual address %016lx\n", (unsigned long)address);
  65         }
  66         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  67                (tsk->mm ?
  68                 CTX_HWBITS(tsk->mm->context) :
  69                 CTX_HWBITS(tsk->active_mm->context)));
  70         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  71                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  72                           (unsigned long) tsk->active_mm->pgd));
  73         die_if_kernel("Oops", regs);
  74 }
  75
  76 static void bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  77 {
  78         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  79                regs->tpc);
  80         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  81         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  82         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  83         dump_stack();
  84         unhandled_fault(regs->tpc, current, regs);
  85 }
  86
  87 /*
  88  * We now make sure that mmap_sem is held in all paths that call
  89  * this. Additionally, to prevent kswapd from ripping ptes from
  90  * under us, raise interrupts around the time that we look at the
  91  * pte, kswapd will have to wait to get his smp ipi response from
  92  * us. vmtruncate likewise. This saves us having to get pte lock.
  93  */
  94 static unsigned int get_user_insn(unsigned long tpc)
  95 {
  96         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  97         pud_t *pudp;
  98         pmd_t *pmdp;
  99         pte_t *ptep, pte;
 100         unsigned long pa;
 101         u32 insn = 0;
 102         unsigned long pstate;
 103
 104         if (pgd_none(*pgdp))
 105                 goto outret;
 106         pudp = pud_offset(pgdp, tpc);
 107         if (pud_none(*pudp))
 108                 goto outret;
 109         pmdp = pmd_offset(pudp, tpc);
 110         if (pmd_none(*pmdp))
 111                 goto outret;
 112
 113         /* This disables preemption for us as well. */
 114         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 115         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 116                                 : : "r" (pstate), "i" (PSTATE_IE));
 117         ptep = pte_offset_map(pmdp, tpc);
 118         pte = *ptep;
 119         if (!pte_present(pte))
 120                 goto out;
 121
 122         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 123         pa += (tpc & ~PAGE_MASK);
 124
 125         /* Use phys bypass so we don't pollute dtlb/dcache. */
 126         __asm__ __volatile__("lduwa [%1] %2, %0"
 127                              : "=r" (insn)
 128                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 129
 130 out:
 131         pte_unmap(ptep);
 132         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 133 outret:
 134         return insn;
 135 }
 136
 137 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 138
 139 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 140                              unsigned int insn, int fault_code)
 141 {
 142         siginfo_t info;
 143
 144         info.si_code = code;
 145         info.si_signo = sig;
 146         info.si_errno = 0;
 147         if (fault_code & FAULT_CODE_ITLB)
 148                 info.si_addr = (void __user *) regs->tpc;
 149         else
 150                 info.si_addr = (void __user *)
 151                         compute_effective_address(regs, insn, 0);
 152         info.si_trapno = 0;
 153         force_sig_info(sig, &info, current);
 154 }
 155
 156 extern int handle_ldf_stq(u32, struct pt_regs *);
 157 extern int handle_ld_nf(u32, struct pt_regs *);
 158
 159 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 160 {
 161         if (!insn) {
 162                 if (!regs->tpc || (regs->tpc & 0x3))
 163                         return 0;
 164                 if (regs->tstate & TSTATE_PRIV) {
 165                         insn = *(unsigned int *) regs->tpc;
 166                 } else {
 167                         insn = get_user_insn(regs->tpc);
 168                 }
 169         }
 170         return insn;
 171 }
 172
 173 static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
 174                             unsigned int insn, unsigned long address)
 175 {
 176         unsigned char asi = ASI_P;
 177
 178         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 179                 goto cannot_handle;
 180
 181         /* If user insn could be read (thus insn is zero), that
 182          * is fine.  We will just gun down the process with a signal
 183          * in that case.
 184          */
 185
 186         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 187             (insn & 0xc0800000) == 0xc0800000) {
 188                 if (insn & 0x2000)
 189                         asi = (regs->tstate >> 24);
 190                 else
 191                         asi = (insn >> 5);
 192                 if ((asi & 0xf2) == 0x82) {
 193                         if (insn & 0x1000000) {
 194                                 handle_ldf_stq(insn, regs);
 195                         } else {
 196                                 /* This was a non-faulting load. Just clear the
 197                                  * destination register(s) and continue with the next
 198                                  * instruction. -jj
 199                                  */
 200                                 handle_ld_nf(insn, regs);
 201                         }
 202                         return;
 203                 }
 204         }
 205
 206         /* Is this in ex_table? */
 207         if (regs->tstate & TSTATE_PRIV) {
 208                 const struct exception_table_entry *entry;
 209
 210                 entry = search_exception_tables(regs->tpc);
 211                 if (entry) {
 212                         regs->tpc = entry->fixup;
 213                         regs->tnpc = regs->tpc + 4;
 214                         return;
 215                 }
 216         } else {
 217                 /* The si_code was set to make clear whether
 218                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 219                  */
 220                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 221                 return;
 222         }
 223
 224 cannot_handle:
 225         unhandled_fault (address, current, regs);
 226 }
 227
 228 static void noinline bogus_32bit_fault_tpc(struct pt_regs *regs)
 229 {
 230         static int times;
 231
 232         if (times++ < 10)
 233                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 234                        "64-bit TPC [%lx]\n",
 235                        current->comm, current->pid,
 236                        regs->tpc);
 237         show_regs(regs);
 238 }
 239
 240 static void noinline bogus_32bit_fault_address(struct pt_regs *regs,
 241                                                unsigned long addr)
 242 {
 243         static int times;
 244
 245         if (times++ < 10)
 246                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process "
 247                        "reports 64-bit fault address [%lx]\n",
 248                        current->comm, current->pid, addr);
 249         show_regs(regs);
 250 }
 251
 252 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 253 {
 254         struct mm_struct *mm = current->mm;
 255         struct vm_area_struct *vma;
 256         unsigned int insn = 0;
 257         int si_code, fault_code, fault;
 258         unsigned long address, mm_rss;
 259
 260         fault_code = get_thread_fault_code();
 261
 262         if (notify_page_fault(regs))
 263                 return;
 264
 265         si_code = SEGV_MAPERR;
 266         address = current_thread_info()->fault_address;
 267
 268         if ((fault_code & FAULT_CODE_ITLB) &&
 269             (fault_code & FAULT_CODE_DTLB))
 270                 BUG();
 271
 272         if (test_thread_flag(TIF_32BIT)) {
 273                 if (!(regs->tstate & TSTATE_PRIV)) {
 274                         if (unlikely((regs->tpc >> 32) != 0)) {
 275                                 bogus_32bit_fault_tpc(regs);
 276                                 goto intr_or_no_mm;
 277                         }
 278                 }
 279                 if (unlikely((address >> 32) != 0)) {
 280                         bogus_32bit_fault_address(regs, address);
 281                         goto intr_or_no_mm;
 282                 }
 283         }
 284
 285         if (regs->tstate & TSTATE_PRIV) {
 286                 unsigned long tpc = regs->tpc;
 287
 288                 /* Sanity check the PC. */
 289                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 290                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 291                         /* Valid, no problems... */
 292                 } else {
 293                         bad_kernel_pc(regs, address);
 294                         return;
 295                 }
 296         }
 297
 298         /*
 299          * If we're in an interrupt or have no user
 300          * context, we must not take the fault..
 301          */
 302         if (in_atomic() || !mm)
 303                 goto intr_or_no_mm;
 304
 305         if (!down_read_trylock(&mm->mmap_sem)) {
 306                 if ((regs->tstate & TSTATE_PRIV) &&
 307                     !search_exception_tables(regs->tpc)) {
 308                         insn = get_fault_insn(regs, insn);
 309                         goto handle_kernel_fault;
 310                 }
 311                 down_read(&mm->mmap_sem);
 312         }
 313
 314         vma = find_vma(mm, address);
 315         if (!vma)
 316                 goto bad_area;
 317
 318         /* Pure DTLB misses do not tell us whether the fault causing
 319          * load/store/atomic was a write or not, it only says that there
 320          * was no match.  So in such a case we (carefully) read the
 321          * instruction to try and figure this out.  It's an optimization
 322          * so it's ok if we can't do this.
 323          *
 324          * Special hack, window spill/fill knows the exact fault type.
 325          */
 326         if (((fault_code &
 327               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 328             (vma->vm_flags & VM_WRITE) != 0) {
 329                 insn = get_fault_insn(regs, 0);
 330                 if (!insn)
 331                         goto continue_fault;
 332                 /* All loads, stores and atomics have bits 30 and 31 both set
 333                  * in the instruction.  Bit 21 is set in all stores, but we
 334                  * have to avoid prefetches which also have bit 21 set.
 335                  */
 336                 if ((insn & 0xc0200000) == 0xc0200000 &&
 337                     (insn & 0x01780000) != 0x01680000) {
 338                         /* Don't bother updating thread struct value,
 339                          * because update_mmu_cache only cares which tlb
 340                          * the access came from.
 341                          */
 342                         fault_code |= FAULT_CODE_WRITE;
 343                 }
 344         }
 345 continue_fault:
 346
 347         if (vma->vm_start <= address)
 348                 goto good_area;
 349         if (!(vma->vm_flags & VM_GROWSDOWN))
 350                 goto bad_area;
 351         if (!(fault_code & FAULT_CODE_WRITE)) {
 352                 /* Non-faulting loads shouldn't expand stack. */
 353                 insn = get_fault_insn(regs, insn);
 354                 if ((insn & 0xc0800000) == 0xc0800000) {
 355                         unsigned char asi;
 356
 357                         if (insn & 0x2000)
 358                                 asi = (regs->tstate >> 24);
 359                         else
 360                                 asi = (insn >> 5);
 361                         if ((asi & 0xf2) == 0x82)
 362                                 goto bad_area;
 363                 }
 364         }
 365         if (expand_stack(vma, address))
 366                 goto bad_area;
 367         /*
 368          * Ok, we have a good vm_area for this memory access, so
 369          * we can handle it..
 370          */
 371 good_area:
 372         si_code = SEGV_ACCERR;
 373
 374         /* If we took a ITLB miss on a non-executable page, catch
 375          * that here.
 376          */
 377         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 378                 BUG_ON(address != regs->tpc);
 379                 BUG_ON(regs->tstate & TSTATE_PRIV);
 380                 goto bad_area;
 381         }
 382
 383         if (fault_code & FAULT_CODE_WRITE) {
 384                 if (!(vma->vm_flags & VM_WRITE))
 385                         goto bad_area;
 386
 387                 /* Spitfire has an icache which does not snoop
 388                  * processor stores.  Later processors do...
 389                  */
 390                 if (tlb_type == spitfire &&
 391                     (vma->vm_flags & VM_EXEC) != 0 &&
 392                     vma->vm_file != NULL)
 393                         set_thread_fault_code(fault_code |
 394                                               FAULT_CODE_BLKCOMMIT);
 395         } else {
 396                 /* Allow reads even for write-only mappings */
 397                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 398                         goto bad_area;
 399         }
 400
 401         fault = handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE) ? FAULT_FLAG_WRITE : 0);
 402         if (unlikely(fault & VM_FAULT_ERROR)) {
 403                 if (fault & VM_FAULT_OOM)
 404                         goto out_of_memory;
 405                 else if (fault & VM_FAULT_SIGBUS)
 406                         goto do_sigbus;
 407                 BUG();
 408         }
 409         if (fault & VM_FAULT_MAJOR)
 410                 current->maj_flt++;
 411         else
 412                 current->min_flt++;
 413
 414         up_read(&mm->mmap_sem);
 415
 416         mm_rss = get_mm_rss(mm);
 417 #ifdef CONFIG_HUGETLB_PAGE
 418         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 419 #endif
 420         if (unlikely(mm_rss >
 421                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 422                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 423 #ifdef CONFIG_HUGETLB_PAGE
 424         mm_rss = mm->context.huge_pte_count;
 425         if (unlikely(mm_rss >
 426                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
 427                 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 428 #endif
 429         return;
 430
 431         /*
 432          * Something tried to access memory that isn't in our memory map..
 433          * Fix it, but check if it's kernel or user first..
 434          */
 435 bad_area:
 436         insn = get_fault_insn(regs, insn);
 437         up_read(&mm->mmap_sem);
 438
 439 handle_kernel_fault:
 440         do_kernel_fault(regs, si_code, fault_code, insn, address);
 441         return;
 442
 443 /*
 444  * We ran out of memory, or some other thing happened to us that made
 445  * us unable to handle the page fault gracefully.
 446  */
 447 out_of_memory:
 448         insn = get_fault_insn(regs, insn);
 449         up_read(&mm->mmap_sem);
 450         if (!(regs->tstate & TSTATE_PRIV)) {
 451                 pagefault_out_of_memory();
 452                 return;
 453         }
 454         goto handle_kernel_fault;
 455
 456 intr_or_no_mm:
 457         insn = get_fault_insn(regs, 0);
 458         goto handle_kernel_fault;
 459
 460 do_sigbus:
 461         insn = get_fault_insn(regs, insn);
 462         up_read(&mm->mmap_sem);
 463
 464         /*
 465          * Send a sigbus, regardless of whether we were in kernel
 466          * or user mode.
 467          */
 468         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 469
 470         /* Kernel mode? Handle exceptions or die */
 471         if (regs->tstate & TSTATE_PRIV)
 472                 goto handle_kernel_fault;
 473 }