arch/sparc64/mm/fault.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
  23 #include <asm/page.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/openprom.h>
  26 #include <asm/oplib.h>
  27 #include <asm/uaccess.h>
  28 #include <asm/asi.h>
  29 #include <asm/lsu.h>
  30 #include <asm/sections.h>
  31 #include <asm/mmu_context.h>
  32
  33 #ifdef CONFIG_KPROBES
  34 static inline int notify_page_fault(struct pt_regs *regs)
  35 {
  36         int ret = 0;
  37
  38         /* kprobe_running() needs smp_processor_id() */
  39         if (!user_mode(regs)) {
  40                 preempt_disable();
  41                 if (kprobe_running() && kprobe_fault_handler(regs, 0))
  42                         ret = 1;
  43                 preempt_enable();
  44         }
  45         return ret;
  46 }
  47 #else
  48 static inline int notify_page_fault(struct pt_regs *regs)
  49 {
  50         return 0;
  51 }
  52 #endif
  53
  54 static void __kprobes unhandled_fault(unsigned long address,
  55                                       struct task_struct *tsk,
  56                                       struct pt_regs *regs)
  57 {
  58         if ((unsigned long) address < PAGE_SIZE) {
  59                 printk(KERN_ALERT "Unable to handle kernel NULL "
  60                        "pointer dereference\n");
  61         } else {
  62                 printk(KERN_ALERT "Unable to handle kernel paging request "
  63                        "at virtual address %016lx\n", (unsigned long)address);
  64         }
  65         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  66                (tsk->mm ?
  67                 CTX_HWBITS(tsk->mm->context) :
  68                 CTX_HWBITS(tsk->active_mm->context)));
  69         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  70                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  71                           (unsigned long) tsk->active_mm->pgd));
  72         die_if_kernel("Oops", regs);
  73 }
  74
  75 static void bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  76 {
  77         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  78                regs->tpc);
  79         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  80         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  81         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  82         dump_stack();
  83         unhandled_fault(regs->tpc, current, regs);
  84 }
  85
  86 /*
  87  * We now make sure that mmap_sem is held in all paths that call
  88  * this. Additionally, to prevent kswapd from ripping ptes from
  89  * under us, raise interrupts around the time that we look at the
  90  * pte, kswapd will have to wait to get his smp ipi response from
  91  * us. vmtruncate likewise. This saves us having to get pte lock.
  92  */
  93 static unsigned int get_user_insn(unsigned long tpc)
  94 {
  95         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  96         pud_t *pudp;
  97         pmd_t *pmdp;
  98         pte_t *ptep, pte;
  99         unsigned long pa;
 100         u32 insn = 0;
 101         unsigned long pstate;
 102
 103         if (pgd_none(*pgdp))
 104                 goto outret;
 105         pudp = pud_offset(pgdp, tpc);
 106         if (pud_none(*pudp))
 107                 goto outret;
 108         pmdp = pmd_offset(pudp, tpc);
 109         if (pmd_none(*pmdp))
 110                 goto outret;
 111
 112         /* This disables preemption for us as well. */
 113         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 114         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 115                                 : : "r" (pstate), "i" (PSTATE_IE));
 116         ptep = pte_offset_map(pmdp, tpc);
 117         pte = *ptep;
 118         if (!pte_present(pte))
 119                 goto out;
 120
 121         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 122         pa += (tpc & ~PAGE_MASK);
 123
 124         /* Use phys bypass so we don't pollute dtlb/dcache. */
 125         __asm__ __volatile__("lduwa [%1] %2, %0"
 126                              : "=r" (insn)
 127                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 128
 129 out:
 130         pte_unmap(ptep);
 131         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 132 outret:
 133         return insn;
 134 }
 135
 136 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 137
 138 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 139                              unsigned int insn, int fault_code)
 140 {
 141         siginfo_t info;
 142
 143         info.si_code = code;
 144         info.si_signo = sig;
 145         info.si_errno = 0;
 146         if (fault_code & FAULT_CODE_ITLB)
 147                 info.si_addr = (void __user *) regs->tpc;
 148         else
 149                 info.si_addr = (void __user *)
 150                         compute_effective_address(regs, insn, 0);
 151         info.si_trapno = 0;
 152         force_sig_info(sig, &info, current);
 153 }
 154
 155 extern int handle_ldf_stq(u32, struct pt_regs *);
 156 extern int handle_ld_nf(u32, struct pt_regs *);
 157
 158 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 159 {
 160         if (!insn) {
 161                 if (!regs->tpc || (regs->tpc & 0x3))
 162                         return 0;
 163                 if (regs->tstate & TSTATE_PRIV) {
 164                         insn = *(unsigned int *) regs->tpc;
 165                 } else {
 166                         insn = get_user_insn(regs->tpc);
 167                 }
 168         }
 169         return insn;
 170 }
 171
 172 static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
 173                             unsigned int insn, unsigned long address)
 174 {
 175         unsigned char asi = ASI_P;
 176
 177         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 178                 goto cannot_handle;
 179
 180         /* If user insn could be read (thus insn is zero), that
 181          * is fine.  We will just gun down the process with a signal
 182          * in that case.
 183          */
 184
 185         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 186             (insn & 0xc0800000) == 0xc0800000) {
 187                 if (insn & 0x2000)
 188                         asi = (regs->tstate >> 24);
 189                 else
 190                         asi = (insn >> 5);
 191                 if ((asi & 0xf2) == 0x82) {
 192                         if (insn & 0x1000000) {
 193                                 handle_ldf_stq(insn, regs);
 194                         } else {
 195                                 /* This was a non-faulting load. Just clear the
 196                                  * destination register(s) and continue with the next
 197                                  * instruction. -jj
 198                                  */
 199                                 handle_ld_nf(insn, regs);
 200                         }
 201                         return;
 202                 }
 203         }
 204
 205         /* Is this in ex_table? */
 206         if (regs->tstate & TSTATE_PRIV) {
 207                 const struct exception_table_entry *entry;
 208
 209                 entry = search_exception_tables(regs->tpc);
 210                 if (entry) {
 211                         regs->tpc = entry->fixup;
 212                         regs->tnpc = regs->tpc + 4;
 213                         return;
 214                 }
 215         } else {
 216                 /* The si_code was set to make clear whether
 217                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 218                  */
 219                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 220                 return;
 221         }
 222
 223 cannot_handle:
 224         unhandled_fault (address, current, regs);
 225 }
 226
 227 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 228 {
 229         struct mm_struct *mm = current->mm;
 230         struct vm_area_struct *vma;
 231         unsigned int insn = 0;
 232         int si_code, fault_code, fault;
 233         unsigned long address, mm_rss;
 234
 235         fault_code = get_thread_fault_code();
 236
 237         if (notify_page_fault(regs))
 238                 return;
 239
 240         si_code = SEGV_MAPERR;
 241         address = current_thread_info()->fault_address;
 242
 243         if ((fault_code & FAULT_CODE_ITLB) &&
 244             (fault_code & FAULT_CODE_DTLB))
 245                 BUG();
 246
 247         if (regs->tstate & TSTATE_PRIV) {
 248                 unsigned long tpc = regs->tpc;
 249
 250                 /* Sanity check the PC. */
 251                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 252                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 253                         /* Valid, no problems... */
 254                 } else {
 255                         bad_kernel_pc(regs, address);
 256                         return;
 257                 }
 258         }
 259
 260         /*
 261          * If we're in an interrupt or have no user
 262          * context, we must not take the fault..
 263          */
 264         if (in_atomic() || !mm)
 265                 goto intr_or_no_mm;
 266
 267         if (test_thread_flag(TIF_32BIT)) {
 268                 if (!(regs->tstate & TSTATE_PRIV))
 269                         regs->tpc &= 0xffffffff;
 270                 address &= 0xffffffff;
 271         }
 272
 273         if (!down_read_trylock(&mm->mmap_sem)) {
 274                 if ((regs->tstate & TSTATE_PRIV) &&
 275                     !search_exception_tables(regs->tpc)) {
 276                         insn = get_fault_insn(regs, insn);
 277                         goto handle_kernel_fault;
 278                 }
 279                 down_read(&mm->mmap_sem);
 280         }
 281
 282         vma = find_vma(mm, address);
 283         if (!vma)
 284                 goto bad_area;
 285
 286         /* Pure DTLB misses do not tell us whether the fault causing
 287          * load/store/atomic was a write or not, it only says that there
 288          * was no match.  So in such a case we (carefully) read the
 289          * instruction to try and figure this out.  It's an optimization
 290          * so it's ok if we can't do this.
 291          *
 292          * Special hack, window spill/fill knows the exact fault type.
 293          */
 294         if (((fault_code &
 295               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 296             (vma->vm_flags & VM_WRITE) != 0) {
 297                 insn = get_fault_insn(regs, 0);
 298                 if (!insn)
 299                         goto continue_fault;
 300                 /* All loads, stores and atomics have bits 30 and 31 both set
 301                  * in the instruction.  Bit 21 is set in all stores, but we
 302                  * have to avoid prefetches which also have bit 21 set.
 303                  */
 304                 if ((insn & 0xc0200000) == 0xc0200000 &&
 305                     (insn & 0x01780000) != 0x01680000) {
 306                         /* Don't bother updating thread struct value,
 307                          * because update_mmu_cache only cares which tlb
 308                          * the access came from.
 309                          */
 310                         fault_code |= FAULT_CODE_WRITE;
 311                 }
 312         }
 313 continue_fault:
 314
 315         if (vma->vm_start <= address)
 316                 goto good_area;
 317         if (!(vma->vm_flags & VM_GROWSDOWN))
 318                 goto bad_area;
 319         if (!(fault_code & FAULT_CODE_WRITE)) {
 320                 /* Non-faulting loads shouldn't expand stack. */
 321                 insn = get_fault_insn(regs, insn);
 322                 if ((insn & 0xc0800000) == 0xc0800000) {
 323                         unsigned char asi;
 324
 325                         if (insn & 0x2000)
 326                                 asi = (regs->tstate >> 24);
 327                         else
 328                                 asi = (insn >> 5);
 329                         if ((asi & 0xf2) == 0x82)
 330                                 goto bad_area;
 331                 }
 332         }
 333         if (expand_stack(vma, address))
 334                 goto bad_area;
 335         /*
 336          * Ok, we have a good vm_area for this memory access, so
 337          * we can handle it..
 338          */
 339 good_area:
 340         si_code = SEGV_ACCERR;
 341
 342         /* If we took a ITLB miss on a non-executable page, catch
 343          * that here.
 344          */
 345         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 346                 BUG_ON(address != regs->tpc);
 347                 BUG_ON(regs->tstate & TSTATE_PRIV);
 348                 goto bad_area;
 349         }
 350
 351         if (fault_code & FAULT_CODE_WRITE) {
 352                 if (!(vma->vm_flags & VM_WRITE))
 353                         goto bad_area;
 354
 355                 /* Spitfire has an icache which does not snoop
 356                  * processor stores.  Later processors do...
 357                  */
 358                 if (tlb_type == spitfire &&
 359                     (vma->vm_flags & VM_EXEC) != 0 &&
 360                     vma->vm_file != NULL)
 361                         set_thread_fault_code(fault_code |
 362                                               FAULT_CODE_BLKCOMMIT);
 363         } else {
 364                 /* Allow reads even for write-only mappings */
 365                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 366                         goto bad_area;
 367         }
 368
 369         fault = handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE));
 370         if (unlikely(fault & VM_FAULT_ERROR)) {
 371                 if (fault & VM_FAULT_OOM)
 372                         goto out_of_memory;
 373                 else if (fault & VM_FAULT_SIGBUS)
 374                         goto do_sigbus;
 375                 BUG();
 376         }
 377         if (fault & VM_FAULT_MAJOR)
 378                 current->maj_flt++;
 379         else
 380                 current->min_flt++;
 381
 382         up_read(&mm->mmap_sem);
 383
 384         mm_rss = get_mm_rss(mm);
 385 #ifdef CONFIG_HUGETLB_PAGE
 386         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 387 #endif
 388         if (unlikely(mm_rss >
 389                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 390                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 391 #ifdef CONFIG_HUGETLB_PAGE
 392         mm_rss = mm->context.huge_pte_count;
 393         if (unlikely(mm_rss >
 394                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
 395                 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 396 #endif
 397         return;
 398
 399         /*
 400          * Something tried to access memory that isn't in our memory map..
 401          * Fix it, but check if it's kernel or user first..
 402          */
 403 bad_area:
 404         insn = get_fault_insn(regs, insn);
 405         up_read(&mm->mmap_sem);
 406
 407 handle_kernel_fault:
 408         do_kernel_fault(regs, si_code, fault_code, insn, address);
 409         return;
 410
 411 /*
 412  * We ran out of memory, or some other thing happened to us that made
 413  * us unable to handle the page fault gracefully.
 414  */
 415 out_of_memory:
 416         insn = get_fault_insn(regs, insn);
 417         up_read(&mm->mmap_sem);
 418         printk("VM: killing process %s\n", current->comm);
 419         if (!(regs->tstate & TSTATE_PRIV))
 420                 do_group_exit(SIGKILL);
 421         goto handle_kernel_fault;
 422
 423 intr_or_no_mm:
 424         insn = get_fault_insn(regs, 0);
 425         goto handle_kernel_fault;
 426
 427 do_sigbus:
 428         insn = get_fault_insn(regs, insn);
 429         up_read(&mm->mmap_sem);
 430
 431         /*
 432          * Send a sigbus, regardless of whether we were in kernel
 433          * or user mode.
 434          */
 435         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 436
 437         /* Kernel mode? Handle exceptions or die */
 438         if (regs->tstate & TSTATE_PRIV)
 439                 goto handle_kernel_fault;
 440 }