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 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
  36 {
  37         int ret = 0;
  38
  39         /* kprobe_running() needs smp_processor_id() */
  40         if (kprobes_built_in() && !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
  49 static void __kprobes unhandled_fault(unsigned long address,
  50                                       struct task_struct *tsk,
  51                                       struct pt_regs *regs)
  52 {
  53         if ((unsigned long) address < PAGE_SIZE) {
  54                 printk(KERN_ALERT "Unable to handle kernel NULL "
  55                        "pointer dereference\n");
  56         } else {
  57                 printk(KERN_ALERT "Unable to handle kernel paging request "
  58                        "at virtual address %016lx\n", (unsigned long)address);
  59         }
  60         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  61                (tsk->mm ?
  62                 CTX_HWBITS(tsk->mm->context) :
  63                 CTX_HWBITS(tsk->active_mm->context)));
  64         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  65                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  66                           (unsigned long) tsk->active_mm->pgd));
  67         die_if_kernel("Oops", regs);
  68 }
  69
  70 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  71 {
  72         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  73                regs->tpc);
  74         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  75         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  76         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  77         dump_stack();
  78         unhandled_fault(regs->tpc, current, regs);
  79 }
  80
  81 /*
  82  * We now make sure that mmap_sem is held in all paths that call
  83  * this. Additionally, to prevent kswapd from ripping ptes from
  84  * under us, raise interrupts around the time that we look at the
  85  * pte, kswapd will have to wait to get his smp ipi response from
  86  * us. vmtruncate likewise. This saves us having to get pte lock.
  87  */
  88 static unsigned int get_user_insn(unsigned long tpc)
  89 {
  90         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  91         pud_t *pudp;
  92         pmd_t *pmdp;
  93         pte_t *ptep, pte;
  94         unsigned long pa;
  95         u32 insn = 0;
  96         unsigned long pstate;
  97
  98         if (pgd_none(*pgdp))
  99                 goto outret;
 100         pudp = pud_offset(pgdp, tpc);
 101         if (pud_none(*pudp))
 102                 goto outret;
 103         pmdp = pmd_offset(pudp, tpc);
 104         if (pmd_none(*pmdp))
 105                 goto outret;
 106
 107         /* This disables preemption for us as well. */
 108         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 109         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 110                                 : : "r" (pstate), "i" (PSTATE_IE));
 111         ptep = pte_offset_map(pmdp, tpc);
 112         pte = *ptep;
 113         if (!pte_present(pte))
 114                 goto out;
 115
 116         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 117         pa += (tpc & ~PAGE_MASK);
 118
 119         /* Use phys bypass so we don't pollute dtlb/dcache. */
 120         __asm__ __volatile__("lduwa [%1] %2, %0"
 121                              : "=r" (insn)
 122                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 123
 124 out:
 125         pte_unmap(ptep);
 126         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 127 outret:
 128         return insn;
 129 }
 130
 131 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 132
 133 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 134                              unsigned int insn, int fault_code)
 135 {
 136         siginfo_t info;
 137
 138         info.si_code = code;
 139         info.si_signo = sig;
 140         info.si_errno = 0;
 141         if (fault_code & FAULT_CODE_ITLB)
 142                 info.si_addr = (void __user *) regs->tpc;
 143         else
 144                 info.si_addr = (void __user *)
 145                         compute_effective_address(regs, insn, 0);
 146         info.si_trapno = 0;
 147         force_sig_info(sig, &info, current);
 148 }
 149
 150 extern int handle_ldf_stq(u32, struct pt_regs *);
 151 extern int handle_ld_nf(u32, struct pt_regs *);
 152
 153 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 154 {
 155         if (!insn) {
 156                 if (!regs->tpc || (regs->tpc & 0x3))
 157                         return 0;
 158                 if (regs->tstate & TSTATE_PRIV) {
 159                         insn = *(unsigned int *) regs->tpc;
 160                 } else {
 161                         insn = get_user_insn(regs->tpc);
 162                 }
 163         }
 164         return insn;
 165 }
 166
 167 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 168                                       int fault_code, unsigned int insn,
 169                                       unsigned long address)
 170 {
 171         unsigned char asi = ASI_P;
 172
 173         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 174                 goto cannot_handle;
 175
 176         /* If user insn could be read (thus insn is zero), that
 177          * is fine.  We will just gun down the process with a signal
 178          * in that case.
 179          */
 180
 181         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 182             (insn & 0xc0800000) == 0xc0800000) {
 183                 if (insn & 0x2000)
 184                         asi = (regs->tstate >> 24);
 185                 else
 186                         asi = (insn >> 5);
 187                 if ((asi & 0xf2) == 0x82) {
 188                         if (insn & 0x1000000) {
 189                                 handle_ldf_stq(insn, regs);
 190                         } else {
 191                                 /* This was a non-faulting load. Just clear the
 192                                  * destination register(s) and continue with the next
 193                                  * instruction. -jj
 194                                  */
 195                                 handle_ld_nf(insn, regs);
 196                         }
 197                         return;
 198                 }
 199         }
 200
 201         /* Is this in ex_table? */
 202         if (regs->tstate & TSTATE_PRIV) {
 203                 const struct exception_table_entry *entry;
 204
 205                 entry = search_exception_tables(regs->tpc);
 206                 if (entry) {
 207                         regs->tpc = entry->fixup;
 208                         regs->tnpc = regs->tpc + 4;
 209                         return;
 210                 }
 211         } else {
 212                 /* The si_code was set to make clear whether
 213                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 214                  */
 215                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 216                 return;
 217         }
 218
 219 cannot_handle:
 220         unhandled_fault (address, current, regs);
 221 }
 222
 223 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 224 {
 225         static int times;
 226
 227         if (times++ < 10)
 228                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 229                        "64-bit TPC [%lx]\n",
 230                        current->comm, current->pid,
 231                        regs->tpc);
 232         show_regs(regs);
 233 }
 234
 235 static void noinline __kprobes bogus_32bit_fault_address(struct pt_regs *regs,
 236                                                          unsigned long addr)
 237 {
 238         static int times;
 239
 240         if (times++ < 10)
 241                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process "
 242                        "reports 64-bit fault address [%lx]\n",
 243                        current->comm, current->pid, addr);
 244         show_regs(regs);
 245 }
 246
 247 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 248 {
 249         struct mm_struct *mm = current->mm;
 250         struct vm_area_struct *vma;
 251         unsigned int insn = 0;
 252         int si_code, fault_code, fault;
 253         unsigned long address, mm_rss;
 254
 255         fault_code = get_thread_fault_code();
 256
 257         if (notify_page_fault(regs))
 258                 return;
 259
 260         si_code = SEGV_MAPERR;
 261         address = current_thread_info()->fault_address;
 262
 263         if ((fault_code & FAULT_CODE_ITLB) &&
 264             (fault_code & FAULT_CODE_DTLB))
 265                 BUG();
 266
 267         if (test_thread_flag(TIF_32BIT)) {
 268                 if (!(regs->tstate & TSTATE_PRIV)) {
 269                         if (unlikely((regs->tpc >> 32) != 0)) {
 270                                 bogus_32bit_fault_tpc(regs);
 271                                 goto intr_or_no_mm;
 272                         }
 273                 }
 274                 if (unlikely((address >> 32) != 0)) {
 275                         bogus_32bit_fault_address(regs, address);
 276                         goto intr_or_no_mm;
 277                 }
 278         }
 279
 280         if (regs->tstate & TSTATE_PRIV) {
 281                 unsigned long tpc = regs->tpc;
 282
 283                 /* Sanity check the PC. */
 284                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 285                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 286                         /* Valid, no problems... */
 287                 } else {
 288                         bad_kernel_pc(regs, address);
 289                         return;
 290                 }
 291         }
 292
 293         /*
 294          * If we're in an interrupt or have no user
 295          * context, we must not take the fault..
 296          */
 297         if (in_atomic() || !mm)
 298                 goto intr_or_no_mm;
 299
 300         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
 301
 302         if (!down_read_trylock(&mm->mmap_sem)) {
 303                 if ((regs->tstate & TSTATE_PRIV) &&
 304                     !search_exception_tables(regs->tpc)) {
 305                         insn = get_fault_insn(regs, insn);
 306                         goto handle_kernel_fault;
 307                 }
 308                 down_read(&mm->mmap_sem);
 309         }
 310
 311         vma = find_vma(mm, address);
 312         if (!vma)
 313                 goto bad_area;
 314
 315         /* Pure DTLB misses do not tell us whether the fault causing
 316          * load/store/atomic was a write or not, it only says that there
 317          * was no match.  So in such a case we (carefully) read the
 318          * instruction to try and figure this out.  It's an optimization
 319          * so it's ok if we can't do this.
 320          *
 321          * Special hack, window spill/fill knows the exact fault type.
 322          */
 323         if (((fault_code &
 324               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 325             (vma->vm_flags & VM_WRITE) != 0) {
 326                 insn = get_fault_insn(regs, 0);
 327                 if (!insn)
 328                         goto continue_fault;
 329                 /* All loads, stores and atomics have bits 30 and 31 both set
 330                  * in the instruction.  Bit 21 is set in all stores, but we
 331                  * have to avoid prefetches which also have bit 21 set.
 332                  */
 333                 if ((insn & 0xc0200000) == 0xc0200000 &&
 334                     (insn & 0x01780000) != 0x01680000) {
 335                         /* Don't bother updating thread struct value,
 336                          * because update_mmu_cache only cares which tlb
 337                          * the access came from.
 338                          */
 339                         fault_code |= FAULT_CODE_WRITE;
 340                 }
 341         }
 342 continue_fault:
 343
 344         if (vma->vm_start <= address)
 345                 goto good_area;
 346         if (!(vma->vm_flags & VM_GROWSDOWN))
 347                 goto bad_area;
 348         if (!(fault_code & FAULT_CODE_WRITE)) {
 349                 /* Non-faulting loads shouldn't expand stack. */
 350                 insn = get_fault_insn(regs, insn);
 351                 if ((insn & 0xc0800000) == 0xc0800000) {
 352                         unsigned char asi;
 353
 354                         if (insn & 0x2000)
 355                                 asi = (regs->tstate >> 24);
 356                         else
 357                                 asi = (insn >> 5);
 358                         if ((asi & 0xf2) == 0x82)
 359                                 goto bad_area;
 360                 }
 361         }
 362         if (expand_stack(vma, address))
 363                 goto bad_area;
 364         /*
 365          * Ok, we have a good vm_area for this memory access, so
 366          * we can handle it..
 367          */
 368 good_area:
 369         si_code = SEGV_ACCERR;
 370
 371         /* If we took a ITLB miss on a non-executable page, catch
 372          * that here.
 373          */
 374         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 375                 BUG_ON(address != regs->tpc);
 376                 BUG_ON(regs->tstate & TSTATE_PRIV);
 377                 goto bad_area;
 378         }
 379
 380         if (fault_code & FAULT_CODE_WRITE) {
 381                 if (!(vma->vm_flags & VM_WRITE))
 382                         goto bad_area;
 383
 384                 /* Spitfire has an icache which does not snoop
 385                  * processor stores.  Later processors do...
 386                  */
 387                 if (tlb_type == spitfire &&
 388                     (vma->vm_flags & VM_EXEC) != 0 &&
 389                     vma->vm_file != NULL)
 390                         set_thread_fault_code(fault_code |
 391                                               FAULT_CODE_BLKCOMMIT);
 392         } else {
 393                 /* Allow reads even for write-only mappings */
 394                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 395                         goto bad_area;
 396         }
 397
 398         fault = handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE) ? FAULT_FLAG_WRITE : 0);
 399         if (unlikely(fault & VM_FAULT_ERROR)) {
 400                 if (fault & VM_FAULT_OOM)
 401                         goto out_of_memory;
 402                 else if (fault & VM_FAULT_SIGBUS)
 403                         goto do_sigbus;
 404                 BUG();
 405         }
 406         if (fault & VM_FAULT_MAJOR) {
 407                 current->maj_flt++;
 408                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
 409                               regs, address);
 410         } else {
 411                 current->min_flt++;
 412                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
 413                               regs, address);
 414         }
 415         up_read(&mm->mmap_sem);
 416
 417         mm_rss = get_mm_rss(mm);
 418 #ifdef CONFIG_HUGETLB_PAGE
 419         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 420 #endif
 421         if (unlikely(mm_rss >
 422                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 423                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 424 #ifdef CONFIG_HUGETLB_PAGE
 425         mm_rss = mm->context.huge_pte_count;
 426         if (unlikely(mm_rss >
 427                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
 428                 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 429 #endif
 430         return;
 431
 432         /*
 433          * Something tried to access memory that isn't in our memory map..
 434          * Fix it, but check if it's kernel or user first..
 435          */
 436 bad_area:
 437         insn = get_fault_insn(regs, insn);
 438         up_read(&mm->mmap_sem);
 439
 440 handle_kernel_fault:
 441         do_kernel_fault(regs, si_code, fault_code, insn, address);
 442         return;
 443
 444 /*
 445  * We ran out of memory, or some other thing happened to us that made
 446  * us unable to handle the page fault gracefully.
 447  */
 448 out_of_memory:
 449         insn = get_fault_insn(regs, insn);
 450         up_read(&mm->mmap_sem);
 451         if (!(regs->tstate & TSTATE_PRIV)) {
 452                 pagefault_out_of_memory();
 453                 return;
 454         }
 455         goto handle_kernel_fault;
 456
 457 intr_or_no_mm:
 458         insn = get_fault_insn(regs, 0);
 459         goto handle_kernel_fault;
 460
 461 do_sigbus:
 462         insn = get_fault_insn(regs, insn);
 463         up_read(&mm->mmap_sem);
 464
 465         /*
 466          * Send a sigbus, regardless of whether we were in kernel
 467          * or user mode.
 468          */
 469         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 470
 471         /* Kernel mode? Handle exceptions or die */
 472         if (regs->tstate & TSTATE_PRIV)
 473                 goto handle_kernel_fault;
 474 }