arch/sparc/mm/fault_64.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
   4  *
   5  * Copyright (C) 1996, 2008 David S. Miller (davem@davemloft.net)
   6  * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
   7  */
   8
   9 #include <asm/head.h>
  10
  11 #include <linux/string.h>
  12 #include <linux/types.h>
  13 #include <linux/sched.h>
  14 #include <linux/sched/debug.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/signal.h>
  18 #include <linux/mm.h>
  19 #include <linux/extable.h>
  20 #include <linux/init.h>
  21 #include <linux/perf_event.h>
  22 #include <linux/interrupt.h>
  23 #include <linux/kprobes.h>
  24 #include <linux/kdebug.h>
  25 #include <linux/percpu.h>
  26 #include <linux/context_tracking.h>
  27 #include <linux/uaccess.h>
  28
  29 #include <asm/page.h>
  30 #include <asm/pgtable.h>
  31 #include <asm/openprom.h>
  32 #include <asm/oplib.h>
  33 #include <asm/asi.h>
  34 #include <asm/lsu.h>
  35 #include <asm/sections.h>
  36 #include <asm/mmu_context.h>
  37 #include <asm/setup.h>
  38
  39 int show_unhandled_signals = 1;
  40
  41 static void __kprobes unhandled_fault(unsigned long address,
  42                                       struct task_struct *tsk,
  43                                       struct pt_regs *regs)
  44 {
  45         if ((unsigned long) address < PAGE_SIZE) {
  46                 printk(KERN_ALERT "Unable to handle kernel NULL "
  47                        "pointer dereference\n");
  48         } else {
  49                 printk(KERN_ALERT "Unable to handle kernel paging request "
  50                        "at virtual address %016lx\n", (unsigned long)address);
  51         }
  52         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  53                (tsk->mm ?
  54                 CTX_HWBITS(tsk->mm->context) :
  55                 CTX_HWBITS(tsk->active_mm->context)));
  56         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  57                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  58                           (unsigned long) tsk->active_mm->pgd));
  59         die_if_kernel("Oops", regs);
  60 }
  61
  62 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  63 {
  64         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
  65                regs->tpc);
  66         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
  67         printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
  68         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
  69         dump_stack();
  70         unhandled_fault(regs->tpc, current, regs);
  71 }
  72
  73 /*
  74  * We now make sure that mmap_sem is held in all paths that call
  75  * this. Additionally, to prevent kswapd from ripping ptes from
  76  * under us, raise interrupts around the time that we look at the
  77  * pte, kswapd will have to wait to get his smp ipi response from
  78  * us. vmtruncate likewise. This saves us having to get pte lock.
  79  */
  80 static unsigned int get_user_insn(unsigned long tpc)
  81 {
  82         pgd_t *pgdp = pgd_offset(current->mm, tpc);
  83         p4d_t *p4dp;
  84         pud_t *pudp;
  85         pmd_t *pmdp;
  86         pte_t *ptep, pte;
  87         unsigned long pa;
  88         u32 insn = 0;
  89
  90         if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
  91                 goto out;
  92         p4dp = p4d_offset(pgdp, tpc);
  93         if (p4d_none(*p4dp) || unlikely(p4d_bad(*p4dp)))
  94                 goto out;
  95         pudp = pud_offset(p4dp, tpc);
  96         if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
  97                 goto out;
  98
  99         /* This disables preemption for us as well. */
 100         local_irq_disable();
 101
 102         pmdp = pmd_offset(pudp, tpc);
 103         if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
 104                 goto out_irq_enable;
 105
 106 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 107         if (is_hugetlb_pmd(*pmdp)) {
 108                 pa  = pmd_pfn(*pmdp) << PAGE_SHIFT;
 109                 pa += tpc & ~HPAGE_MASK;
 110
 111                 /* Use phys bypass so we don't pollute dtlb/dcache. */
 112                 __asm__ __volatile__("lduwa [%1] %2, %0"
 113                                      : "=r" (insn)
 114                                      : "r" (pa), "i" (ASI_PHYS_USE_EC));
 115         } else
 116 #endif
 117         {
 118                 ptep = pte_offset_map(pmdp, tpc);
 119                 pte = *ptep;
 120                 if (pte_present(pte)) {
 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                 pte_unmap(ptep);
 130         }
 131 out_irq_enable:
 132         local_irq_enable();
 133 out:
 134         return insn;
 135 }
 136
 137 static inline void
 138 show_signal_msg(struct pt_regs *regs, int sig, int code,
 139                 unsigned long address, struct task_struct *tsk)
 140 {
 141         if (!unhandled_signal(tsk, sig))
 142                 return;
 143
 144         if (!printk_ratelimit())
 145                 return;
 146
 147         printk("%s%s[%d]: segfault at %lx ip %px (rpc %px) sp %px error %x",
 148                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 149                tsk->comm, task_pid_nr(tsk), address,
 150                (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
 151                (void *)regs->u_regs[UREG_FP], code);
 152
 153         print_vma_addr(KERN_CONT " in ", regs->tpc);
 154
 155         printk(KERN_CONT "\n");
 156 }
 157
 158 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 159                              unsigned long fault_addr, unsigned int insn,
 160                              int fault_code)
 161 {
 162         unsigned long addr;
 163
 164         if (fault_code & FAULT_CODE_ITLB) {
 165                 addr = regs->tpc;
 166         } else {
 167                 /* If we were able to probe the faulting instruction, use it
 168                  * to compute a precise fault address.  Otherwise use the fault
 169                  * time provided address which may only have page granularity.
 170                  */
 171                 if (insn)
 172                         addr = compute_effective_address(regs, insn, 0);
 173                 else
 174                         addr = fault_addr;
 175         }
 176
 177         if (unlikely(show_unhandled_signals))
 178                 show_signal_msg(regs, sig, code, addr, current);
 179
 180         force_sig_fault(sig, code, (void __user *) addr, 0);
 181 }
 182
 183 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 184 {
 185         if (!insn) {
 186                 if (!regs->tpc || (regs->tpc & 0x3))
 187                         return 0;
 188                 if (regs->tstate & TSTATE_PRIV) {
 189                         insn = *(unsigned int *) regs->tpc;
 190                 } else {
 191                         insn = get_user_insn(regs->tpc);
 192                 }
 193         }
 194         return insn;
 195 }
 196
 197 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 198                                       int fault_code, unsigned int insn,
 199                                       unsigned long address)
 200 {
 201         unsigned char asi = ASI_P;
 202
 203         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 204                 goto cannot_handle;
 205
 206         /* If user insn could be read (thus insn is zero), that
 207          * is fine.  We will just gun down the process with a signal
 208          * in that case.
 209          */
 210
 211         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 212             (insn & 0xc0800000) == 0xc0800000) {
 213                 if (insn & 0x2000)
 214                         asi = (regs->tstate >> 24);
 215                 else
 216                         asi = (insn >> 5);
 217                 if ((asi & 0xf2) == 0x82) {
 218                         if (insn & 0x1000000) {
 219                                 handle_ldf_stq(insn, regs);
 220                         } else {
 221                                 /* This was a non-faulting load. Just clear the
 222                                  * destination register(s) and continue with the next
 223                                  * instruction. -jj
 224                                  */
 225                                 handle_ld_nf(insn, regs);
 226                         }
 227                         return;
 228                 }
 229         }
 230
 231         /* Is this in ex_table? */
 232         if (regs->tstate & TSTATE_PRIV) {
 233                 const struct exception_table_entry *entry;
 234
 235                 entry = search_exception_tables(regs->tpc);
 236                 if (entry) {
 237                         regs->tpc = entry->fixup;
 238                         regs->tnpc = regs->tpc + 4;
 239                         return;
 240                 }
 241         } else {
 242                 /* The si_code was set to make clear whether
 243                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 244                  */
 245                 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
 246                 return;
 247         }
 248
 249 cannot_handle:
 250         unhandled_fault (address, current, regs);
 251 }
 252
 253 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 254 {
 255         static int times;
 256
 257         if (times++ < 10)
 258                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 259                        "64-bit TPC [%lx]\n",
 260                        current->comm, current->pid,
 261                        regs->tpc);
 262         show_regs(regs);
 263 }
 264
 265 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 266 {
 267         enum ctx_state prev_state = exception_enter();
 268         struct mm_struct *mm = current->mm;
 269         struct vm_area_struct *vma;
 270         unsigned int insn = 0;
 271         int si_code, fault_code;
 272         vm_fault_t fault;
 273         unsigned long address, mm_rss;
 274         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 275
 276         fault_code = get_thread_fault_code();
 277
 278         if (kprobe_page_fault(regs, 0))
 279                 goto exit_exception;
 280
 281         si_code = SEGV_MAPERR;
 282         address = current_thread_info()->fault_address;
 283
 284         if ((fault_code & FAULT_CODE_ITLB) &&
 285             (fault_code & FAULT_CODE_DTLB))
 286                 BUG();
 287
 288         if (test_thread_flag(TIF_32BIT)) {
 289                 if (!(regs->tstate & TSTATE_PRIV)) {
 290                         if (unlikely((regs->tpc >> 32) != 0)) {
 291                                 bogus_32bit_fault_tpc(regs);
 292                                 goto intr_or_no_mm;
 293                         }
 294                 }
 295                 if (unlikely((address >> 32) != 0))
 296                         goto intr_or_no_mm;
 297         }
 298
 299         if (regs->tstate & TSTATE_PRIV) {
 300                 unsigned long tpc = regs->tpc;
 301
 302                 /* Sanity check the PC. */
 303                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 304                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 305                         /* Valid, no problems... */
 306                 } else {
 307                         bad_kernel_pc(regs, address);
 308                         goto exit_exception;
 309                 }
 310         } else
 311                 flags |= FAULT_FLAG_USER;
 312
 313         /*
 314          * If we're in an interrupt or have no user
 315          * context, we must not take the fault..
 316          */
 317         if (faulthandler_disabled() || !mm)
 318                 goto intr_or_no_mm;
 319
 320         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 321
 322         if (!down_read_trylock(&mm->mmap_sem)) {
 323                 if ((regs->tstate & TSTATE_PRIV) &&
 324                     !search_exception_tables(regs->tpc)) {
 325                         insn = get_fault_insn(regs, insn);
 326                         goto handle_kernel_fault;
 327                 }
 328
 329 retry:
 330                 down_read(&mm->mmap_sem);
 331         }
 332
 333         if (fault_code & FAULT_CODE_BAD_RA)
 334                 goto do_sigbus;
 335
 336         vma = find_vma(mm, address);
 337         if (!vma)
 338                 goto bad_area;
 339
 340         /* Pure DTLB misses do not tell us whether the fault causing
 341          * load/store/atomic was a write or not, it only says that there
 342          * was no match.  So in such a case we (carefully) read the
 343          * instruction to try and figure this out.  It's an optimization
 344          * so it's ok if we can't do this.
 345          *
 346          * Special hack, window spill/fill knows the exact fault type.
 347          */
 348         if (((fault_code &
 349               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 350             (vma->vm_flags & VM_WRITE) != 0) {
 351                 insn = get_fault_insn(regs, 0);
 352                 if (!insn)
 353                         goto continue_fault;
 354                 /* All loads, stores and atomics have bits 30 and 31 both set
 355                  * in the instruction.  Bit 21 is set in all stores, but we
 356                  * have to avoid prefetches which also have bit 21 set.
 357                  */
 358                 if ((insn & 0xc0200000) == 0xc0200000 &&
 359                     (insn & 0x01780000) != 0x01680000) {
 360                         /* Don't bother updating thread struct value,
 361                          * because update_mmu_cache only cares which tlb
 362                          * the access came from.
 363                          */
 364                         fault_code |= FAULT_CODE_WRITE;
 365                 }
 366         }
 367 continue_fault:
 368
 369         if (vma->vm_start <= address)
 370                 goto good_area;
 371         if (!(vma->vm_flags & VM_GROWSDOWN))
 372                 goto bad_area;
 373         if (!(fault_code & FAULT_CODE_WRITE)) {
 374                 /* Non-faulting loads shouldn't expand stack. */
 375                 insn = get_fault_insn(regs, insn);
 376                 if ((insn & 0xc0800000) == 0xc0800000) {
 377                         unsigned char asi;
 378
 379                         if (insn & 0x2000)
 380                                 asi = (regs->tstate >> 24);
 381                         else
 382                                 asi = (insn >> 5);
 383                         if ((asi & 0xf2) == 0x82)
 384                                 goto bad_area;
 385                 }
 386         }
 387         if (expand_stack(vma, address))
 388                 goto bad_area;
 389         /*
 390          * Ok, we have a good vm_area for this memory access, so
 391          * we can handle it..
 392          */
 393 good_area:
 394         si_code = SEGV_ACCERR;
 395
 396         /* If we took a ITLB miss on a non-executable page, catch
 397          * that here.
 398          */
 399         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 400                 WARN(address != regs->tpc,
 401                      "address (%lx) != regs->tpc (%lx)\n", address, regs->tpc);
 402                 WARN_ON(regs->tstate & TSTATE_PRIV);
 403                 goto bad_area;
 404         }
 405
 406         if (fault_code & FAULT_CODE_WRITE) {
 407                 if (!(vma->vm_flags & VM_WRITE))
 408                         goto bad_area;
 409
 410                 /* Spitfire has an icache which does not snoop
 411                  * processor stores.  Later processors do...
 412                  */
 413                 if (tlb_type == spitfire &&
 414                     (vma->vm_flags & VM_EXEC) != 0 &&
 415                     vma->vm_file != NULL)
 416                         set_thread_fault_code(fault_code |
 417                                               FAULT_CODE_BLKCOMMIT);
 418
 419                 flags |= FAULT_FLAG_WRITE;
 420         } else {
 421                 /* Allow reads even for write-only mappings */
 422                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 423                         goto bad_area;
 424         }
 425
 426         fault = handle_mm_fault(vma, address, flags);
 427
 428         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 429                 goto exit_exception;
 430
 431         if (unlikely(fault & VM_FAULT_ERROR)) {
 432                 if (fault & VM_FAULT_OOM)
 433                         goto out_of_memory;
 434                 else if (fault & VM_FAULT_SIGSEGV)
 435                         goto bad_area;
 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_TRANSPARENT_HUGEPAGE)
 467         mm_rss -= (mm->context.thp_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.hugetlb_pte_count + mm->context.thp_pte_count;
 474         mm_rss *= REAL_HPAGE_PER_HPAGE;
 475         if (unlikely(mm_rss >
 476                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
 477                 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
 478                         tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 479                 else
 480                         hugetlb_setup(regs);
 481
 482         }
 483 #endif
 484 exit_exception:
 485         exception_exit(prev_state);
 486         return;
 487
 488         /*
 489          * Something tried to access memory that isn't in our memory map..
 490          * Fix it, but check if it's kernel or user first..
 491          */
 492 bad_area:
 493         insn = get_fault_insn(regs, insn);
 494         up_read(&mm->mmap_sem);
 495
 496 handle_kernel_fault:
 497         do_kernel_fault(regs, si_code, fault_code, insn, address);
 498         goto exit_exception;
 499
 500 /*
 501  * We ran out of memory, or some other thing happened to us that made
 502  * us unable to handle the page fault gracefully.
 503  */
 504 out_of_memory:
 505         insn = get_fault_insn(regs, insn);
 506         up_read(&mm->mmap_sem);
 507         if (!(regs->tstate & TSTATE_PRIV)) {
 508                 pagefault_out_of_memory();
 509                 goto exit_exception;
 510         }
 511         goto handle_kernel_fault;
 512
 513 intr_or_no_mm:
 514         insn = get_fault_insn(regs, 0);
 515         goto handle_kernel_fault;
 516
 517 do_sigbus:
 518         insn = get_fault_insn(regs, insn);
 519         up_read(&mm->mmap_sem);
 520
 521         /*
 522          * Send a sigbus, regardless of whether we were in kernel
 523          * or user mode.
 524          */
 525         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
 526
 527         /* Kernel mode? Handle exceptions or die */
 528         if (regs->tstate & TSTATE_PRIV)
 529                 goto handle_kernel_fault;
 530 }