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 /*
  55  * To debug kernel to catch accesses to certain virtual/physical addresses.
  56  * Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints.
  57  * flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses.
  58  * Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be
  59  * watched. This is only useful on a single cpu machine for now. After the watchpoint
  60  * is detected, the process causing it will be killed, thus preventing an infinite loop.
  61  */
  62 void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode)
  63 {
  64         unsigned long lsubits;
  65
  66         __asm__ __volatile__("ldxa [%%g0] %1, %0"
  67                              : "=r" (lsubits)
  68                              : "i" (ASI_LSU_CONTROL));
  69         lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM |
  70                      LSU_CONTROL_PR | LSU_CONTROL_VR |
  71                      LSU_CONTROL_PW | LSU_CONTROL_VW);
  72
  73         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
  74                              "membar    #Sync"
  75                              : /* no outputs */
  76                              : "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT),
  77                                "i" (ASI_DMMU));
  78
  79         lsubits |= ((unsigned long)mask << (mode ? 25 : 33));
  80         if (flags & VM_READ)
  81                 lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR);
  82         if (flags & VM_WRITE)
  83                 lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW);
  84         __asm__ __volatile__("stxa %0, [%%g0] %1\n\t"
  85                              "membar #Sync"
  86                              : /* no outputs */
  87                              : "r" (lsubits), "i" (ASI_LSU_CONTROL)
  88                              : "memory");
  89 }
  90
  91 static void __kprobes unhandled_fault(unsigned long address,
  92                                       struct task_struct *tsk,
  93                                       struct pt_regs *regs)
  94 {
  95         if ((unsigned long) address < PAGE_SIZE) {
  96                 printk(KERN_ALERT "Unable to handle kernel NULL "
  97                        "pointer dereference\n");
  98         } else {
  99                 printk(KERN_ALERT "Unable to handle kernel paging request "
 100                        "at virtual address %016lx\n", (unsigned long)address);
 101         }
 102         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
 103                (tsk->mm ?
 104                 CTX_HWBITS(tsk->mm->context) :
 105                 CTX_HWBITS(tsk->active_mm->context)));
 106         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
 107                (tsk->mm ? (unsigned long) tsk->mm->pgd :
 108                           (unsigned long) tsk->active_mm->pgd));
 109         die_if_kernel("Oops", regs);
 110 }
 111
 112 static void bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
 113 {
 114         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
 115                regs->tpc);
 116         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
 117         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
 118         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
 119         dump_stack();
 120         unhandled_fault(regs->tpc, current, regs);
 121 }
 122
 123 /*
 124  * We now make sure that mmap_sem is held in all paths that call
 125  * this. Additionally, to prevent kswapd from ripping ptes from
 126  * under us, raise interrupts around the time that we look at the
 127  * pte, kswapd will have to wait to get his smp ipi response from
 128  * us. vmtruncate likewise. This saves us having to get pte lock.
 129  */
 130 static unsigned int get_user_insn(unsigned long tpc)
 131 {
 132         pgd_t *pgdp = pgd_offset(current->mm, tpc);
 133         pud_t *pudp;
 134         pmd_t *pmdp;
 135         pte_t *ptep, pte;
 136         unsigned long pa;
 137         u32 insn = 0;
 138         unsigned long pstate;
 139
 140         if (pgd_none(*pgdp))
 141                 goto outret;
 142         pudp = pud_offset(pgdp, tpc);
 143         if (pud_none(*pudp))
 144                 goto outret;
 145         pmdp = pmd_offset(pudp, tpc);
 146         if (pmd_none(*pmdp))
 147                 goto outret;
 148
 149         /* This disables preemption for us as well. */
 150         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 151         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 152                                 : : "r" (pstate), "i" (PSTATE_IE));
 153         ptep = pte_offset_map(pmdp, tpc);
 154         pte = *ptep;
 155         if (!pte_present(pte))
 156                 goto out;
 157
 158         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 159         pa += (tpc & ~PAGE_MASK);
 160
 161         /* Use phys bypass so we don't pollute dtlb/dcache. */
 162         __asm__ __volatile__("lduwa [%1] %2, %0"
 163                              : "=r" (insn)
 164                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 165
 166 out:
 167         pte_unmap(ptep);
 168         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 169 outret:
 170         return insn;
 171 }
 172
 173 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 174
 175 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 176                              unsigned int insn, int fault_code)
 177 {
 178         siginfo_t info;
 179
 180         info.si_code = code;
 181         info.si_signo = sig;
 182         info.si_errno = 0;
 183         if (fault_code & FAULT_CODE_ITLB)
 184                 info.si_addr = (void __user *) regs->tpc;
 185         else
 186                 info.si_addr = (void __user *)
 187                         compute_effective_address(regs, insn, 0);
 188         info.si_trapno = 0;
 189         force_sig_info(sig, &info, current);
 190 }
 191
 192 extern int handle_ldf_stq(u32, struct pt_regs *);
 193 extern int handle_ld_nf(u32, struct pt_regs *);
 194
 195 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 196 {
 197         if (!insn) {
 198                 if (!regs->tpc || (regs->tpc & 0x3))
 199                         return 0;
 200                 if (regs->tstate & TSTATE_PRIV) {
 201                         insn = *(unsigned int *) regs->tpc;
 202                 } else {
 203                         insn = get_user_insn(regs->tpc);
 204                 }
 205         }
 206         return insn;
 207 }
 208
 209 static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
 210                             unsigned int insn, unsigned long address)
 211 {
 212         unsigned char asi = ASI_P;
 213
 214         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 215                 goto cannot_handle;
 216
 217         /* If user insn could be read (thus insn is zero), that
 218          * is fine.  We will just gun down the process with a signal
 219          * in that case.
 220          */
 221
 222         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 223             (insn & 0xc0800000) == 0xc0800000) {
 224                 if (insn & 0x2000)
 225                         asi = (regs->tstate >> 24);
 226                 else
 227                         asi = (insn >> 5);
 228                 if ((asi & 0xf2) == 0x82) {
 229                         if (insn & 0x1000000) {
 230                                 handle_ldf_stq(insn, regs);
 231                         } else {
 232                                 /* This was a non-faulting load. Just clear the
 233                                  * destination register(s) and continue with the next
 234                                  * instruction. -jj
 235                                  */
 236                                 handle_ld_nf(insn, regs);
 237                         }
 238                         return;
 239                 }
 240         }
 241
 242         /* Is this in ex_table? */
 243         if (regs->tstate & TSTATE_PRIV) {
 244                 const struct exception_table_entry *entry;
 245
 246                 entry = search_exception_tables(regs->tpc);
 247                 if (entry) {
 248                         regs->tpc = entry->fixup;
 249                         regs->tnpc = regs->tpc + 4;
 250                         return;
 251                 }
 252         } else {
 253                 /* The si_code was set to make clear whether
 254                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 255                  */
 256                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 257                 return;
 258         }
 259
 260 cannot_handle:
 261         unhandled_fault (address, current, regs);
 262 }
 263
 264 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 265 {
 266         struct mm_struct *mm = current->mm;
 267         struct vm_area_struct *vma;
 268         unsigned int insn = 0;
 269         int si_code, fault_code, fault;
 270         unsigned long address, mm_rss;
 271
 272         fault_code = get_thread_fault_code();
 273
 274         if (notify_page_fault(regs))
 275                 return;
 276
 277         si_code = SEGV_MAPERR;
 278         address = current_thread_info()->fault_address;
 279
 280         if ((fault_code & FAULT_CODE_ITLB) &&
 281             (fault_code & FAULT_CODE_DTLB))
 282                 BUG();
 283
 284         if (regs->tstate & TSTATE_PRIV) {
 285                 unsigned long tpc = regs->tpc;
 286
 287                 /* Sanity check the PC. */
 288                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 289                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 290                         /* Valid, no problems... */
 291                 } else {
 292                         bad_kernel_pc(regs, address);
 293                         return;
 294                 }
 295         }
 296
 297         /*
 298          * If we're in an interrupt or have no user
 299          * context, we must not take the fault..
 300          */
 301         if (in_atomic() || !mm)
 302                 goto intr_or_no_mm;
 303
 304         if (test_thread_flag(TIF_32BIT)) {
 305                 if (!(regs->tstate & TSTATE_PRIV))
 306                         regs->tpc &= 0xffffffff;
 307                 address &= 0xffffffff;
 308         }
 309
 310         if (!down_read_trylock(&mm->mmap_sem)) {
 311                 if ((regs->tstate & TSTATE_PRIV) &&
 312                     !search_exception_tables(regs->tpc)) {
 313                         insn = get_fault_insn(regs, insn);
 314                         goto handle_kernel_fault;
 315                 }
 316                 down_read(&mm->mmap_sem);
 317         }
 318
 319         vma = find_vma(mm, address);
 320         if (!vma)
 321                 goto bad_area;
 322
 323         /* Pure DTLB misses do not tell us whether the fault causing
 324          * load/store/atomic was a write or not, it only says that there
 325          * was no match.  So in such a case we (carefully) read the
 326          * instruction to try and figure this out.  It's an optimization
 327          * so it's ok if we can't do this.
 328          *
 329          * Special hack, window spill/fill knows the exact fault type.
 330          */
 331         if (((fault_code &
 332               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 333             (vma->vm_flags & VM_WRITE) != 0) {
 334                 insn = get_fault_insn(regs, 0);
 335                 if (!insn)
 336                         goto continue_fault;
 337                 /* All loads, stores and atomics have bits 30 and 31 both set
 338                  * in the instruction.  Bit 21 is set in all stores, but we
 339                  * have to avoid prefetches which also have bit 21 set.
 340                  */
 341                 if ((insn & 0xc0200000) == 0xc0200000 &&
 342                     (insn & 0x01780000) != 0x01680000) {
 343                         /* Don't bother updating thread struct value,
 344                          * because update_mmu_cache only cares which tlb
 345                          * the access came from.
 346                          */
 347                         fault_code |= FAULT_CODE_WRITE;
 348                 }
 349         }
 350 continue_fault:
 351
 352         if (vma->vm_start <= address)
 353                 goto good_area;
 354         if (!(vma->vm_flags & VM_GROWSDOWN))
 355                 goto bad_area;
 356         if (!(fault_code & FAULT_CODE_WRITE)) {
 357                 /* Non-faulting loads shouldn't expand stack. */
 358                 insn = get_fault_insn(regs, insn);
 359                 if ((insn & 0xc0800000) == 0xc0800000) {
 360                         unsigned char asi;
 361
 362                         if (insn & 0x2000)
 363                                 asi = (regs->tstate >> 24);
 364                         else
 365                                 asi = (insn >> 5);
 366                         if ((asi & 0xf2) == 0x82)
 367                                 goto bad_area;
 368                 }
 369         }
 370         if (expand_stack(vma, address))
 371                 goto bad_area;
 372         /*
 373          * Ok, we have a good vm_area for this memory access, so
 374          * we can handle it..
 375          */
 376 good_area:
 377         si_code = SEGV_ACCERR;
 378
 379         /* If we took a ITLB miss on a non-executable page, catch
 380          * that here.
 381          */
 382         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 383                 BUG_ON(address != regs->tpc);
 384                 BUG_ON(regs->tstate & TSTATE_PRIV);
 385                 goto bad_area;
 386         }
 387
 388         if (fault_code & FAULT_CODE_WRITE) {
 389                 if (!(vma->vm_flags & VM_WRITE))
 390                         goto bad_area;
 391
 392                 /* Spitfire has an icache which does not snoop
 393                  * processor stores.  Later processors do...
 394                  */
 395                 if (tlb_type == spitfire &&
 396                     (vma->vm_flags & VM_EXEC) != 0 &&
 397                     vma->vm_file != NULL)
 398                         set_thread_fault_code(fault_code |
 399                                               FAULT_CODE_BLKCOMMIT);
 400         } else {
 401                 /* Allow reads even for write-only mappings */
 402                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 403                         goto bad_area;
 404         }
 405
 406         fault = handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE));
 407         if (unlikely(fault & VM_FAULT_ERROR)) {
 408                 if (fault & VM_FAULT_OOM)
 409                         goto out_of_memory;
 410                 else if (fault & VM_FAULT_SIGBUS)
 411                         goto do_sigbus;
 412                 BUG();
 413         }
 414         if (fault & VM_FAULT_MAJOR)
 415                 current->maj_flt++;
 416         else
 417                 current->min_flt++;
 418
 419         up_read(&mm->mmap_sem);
 420
 421         mm_rss = get_mm_rss(mm);
 422 #ifdef CONFIG_HUGETLB_PAGE
 423         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 424 #endif
 425         if (unlikely(mm_rss >
 426                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 427                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 428 #ifdef CONFIG_HUGETLB_PAGE
 429         mm_rss = mm->context.huge_pte_count;
 430         if (unlikely(mm_rss >
 431                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
 432                 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 433 #endif
 434         return;
 435
 436         /*
 437          * Something tried to access memory that isn't in our memory map..
 438          * Fix it, but check if it's kernel or user first..
 439          */
 440 bad_area:
 441         insn = get_fault_insn(regs, insn);
 442         up_read(&mm->mmap_sem);
 443
 444 handle_kernel_fault:
 445         do_kernel_fault(regs, si_code, fault_code, insn, address);
 446         return;
 447
 448 /*
 449  * We ran out of memory, or some other thing happened to us that made
 450  * us unable to handle the page fault gracefully.
 451  */
 452 out_of_memory:
 453         insn = get_fault_insn(regs, insn);
 454         up_read(&mm->mmap_sem);
 455         printk("VM: killing process %s\n", current->comm);
 456         if (!(regs->tstate & TSTATE_PRIV))
 457                 do_group_exit(SIGKILL);
 458         goto handle_kernel_fault;
 459
 460 intr_or_no_mm:
 461         insn = get_fault_insn(regs, 0);
 462         goto handle_kernel_fault;
 463
 464 do_sigbus:
 465         insn = get_fault_insn(regs, insn);
 466         up_read(&mm->mmap_sem);
 467
 468         /*
 469          * Send a sigbus, regardless of whether we were in kernel
 470          * or user mode.
 471          */
 472         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 473
 474         /* Kernel mode? Handle exceptions or die */
 475         if (regs->tstate & TSTATE_PRIV)
 476                 goto handle_kernel_fault;
 477 }