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 inline __kprobes int notify_page_fault(struct pt_regs *regs)
  42 {
  43         int ret = 0;
  44
  45         /* kprobe_running() needs smp_processor_id() */
  46         if (kprobes_built_in() && !user_mode(regs)) {
  47                 preempt_disable();
  48                 if (kprobe_running() && kprobe_fault_handler(regs, 0))
  49                         ret = 1;
  50                 preempt_enable();
  51         }
  52         return ret;
  53 }
  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 __kprobes 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
 103         if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
 104                 goto out;
 105         pudp = pud_offset(pgdp, tpc);
 106         if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
 107                 goto out;
 108
 109         /* This disables preemption for us as well. */
 110         local_irq_disable();
 111
 112         pmdp = pmd_offset(pudp, tpc);
 113         if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
 114                 goto out_irq_enable;
 115
 116 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 117         if (is_hugetlb_pmd(*pmdp)) {
 118                 pa  = pmd_pfn(*pmdp) << PAGE_SHIFT;
 119                 pa += tpc & ~HPAGE_MASK;
 120
 121                 /* Use phys bypass so we don't pollute dtlb/dcache. */
 122                 __asm__ __volatile__("lduwa [%1] %2, %0"
 123                                      : "=r" (insn)
 124                                      : "r" (pa), "i" (ASI_PHYS_USE_EC));
 125         } else
 126 #endif
 127         {
 128                 ptep = pte_offset_map(pmdp, tpc);
 129                 pte = *ptep;
 130                 if (pte_present(pte)) {
 131                         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 132                         pa += (tpc & ~PAGE_MASK);
 133
 134                         /* Use phys bypass so we don't pollute dtlb/dcache. */
 135                         __asm__ __volatile__("lduwa [%1] %2, %0"
 136                                              : "=r" (insn)
 137                                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 138                 }
 139                 pte_unmap(ptep);
 140         }
 141 out_irq_enable:
 142         local_irq_enable();
 143 out:
 144         return insn;
 145 }
 146
 147 static inline void
 148 show_signal_msg(struct pt_regs *regs, int sig, int code,
 149                 unsigned long address, struct task_struct *tsk)
 150 {
 151         if (!unhandled_signal(tsk, sig))
 152                 return;
 153
 154         if (!printk_ratelimit())
 155                 return;
 156
 157         printk("%s%s[%d]: segfault at %lx ip %px (rpc %px) sp %px error %x",
 158                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 159                tsk->comm, task_pid_nr(tsk), address,
 160                (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
 161                (void *)regs->u_regs[UREG_FP], code);
 162
 163         print_vma_addr(KERN_CONT " in ", regs->tpc);
 164
 165         printk(KERN_CONT "\n");
 166 }
 167
 168 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 169                              unsigned long fault_addr, unsigned int insn,
 170                              int fault_code)
 171 {
 172         unsigned long addr;
 173
 174         if (fault_code & FAULT_CODE_ITLB) {
 175                 addr = regs->tpc;
 176         } else {
 177                 /* If we were able to probe the faulting instruction, use it
 178                  * to compute a precise fault address.  Otherwise use the fault
 179                  * time provided address which may only have page granularity.
 180                  */
 181                 if (insn)
 182                         addr = compute_effective_address(regs, insn, 0);
 183                 else
 184                         addr = fault_addr;
 185         }
 186
 187         if (unlikely(show_unhandled_signals))
 188                 show_signal_msg(regs, sig, code, addr, current);
 189
 190         force_sig_fault(sig, code, (void __user *) addr, 0, current);
 191 }
 192
 193 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 194 {
 195         if (!insn) {
 196                 if (!regs->tpc || (regs->tpc & 0x3))
 197                         return 0;
 198                 if (regs->tstate & TSTATE_PRIV) {
 199                         insn = *(unsigned int *) regs->tpc;
 200                 } else {
 201                         insn = get_user_insn(regs->tpc);
 202                 }
 203         }
 204         return insn;
 205 }
 206
 207 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
 208                                       int fault_code, unsigned int insn,
 209                                       unsigned long address)
 210 {
 211         unsigned char asi = ASI_P;
 212
 213         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 214                 goto cannot_handle;
 215
 216         /* If user insn could be read (thus insn is zero), that
 217          * is fine.  We will just gun down the process with a signal
 218          * in that case.
 219          */
 220
 221         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 222             (insn & 0xc0800000) == 0xc0800000) {
 223                 if (insn & 0x2000)
 224                         asi = (regs->tstate >> 24);
 225                 else
 226                         asi = (insn >> 5);
 227                 if ((asi & 0xf2) == 0x82) {
 228                         if (insn & 0x1000000) {
 229                                 handle_ldf_stq(insn, regs);
 230                         } else {
 231                                 /* This was a non-faulting load. Just clear the
 232                                  * destination register(s) and continue with the next
 233                                  * instruction. -jj
 234                                  */
 235                                 handle_ld_nf(insn, regs);
 236                         }
 237                         return;
 238                 }
 239         }
 240
 241         /* Is this in ex_table? */
 242         if (regs->tstate & TSTATE_PRIV) {
 243                 const struct exception_table_entry *entry;
 244
 245                 entry = search_exception_tables(regs->tpc);
 246                 if (entry) {
 247                         regs->tpc = entry->fixup;
 248                         regs->tnpc = regs->tpc + 4;
 249                         return;
 250                 }
 251         } else {
 252                 /* The si_code was set to make clear whether
 253                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 254                  */
 255                 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
 256                 return;
 257         }
 258
 259 cannot_handle:
 260         unhandled_fault (address, current, regs);
 261 }
 262
 263 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
 264 {
 265         static int times;
 266
 267         if (times++ < 10)
 268                 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
 269                        "64-bit TPC [%lx]\n",
 270                        current->comm, current->pid,
 271                        regs->tpc);
 272         show_regs(regs);
 273 }
 274
 275 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 276 {
 277         enum ctx_state prev_state = exception_enter();
 278         struct mm_struct *mm = current->mm;
 279         struct vm_area_struct *vma;
 280         unsigned int insn = 0;
 281         int si_code, fault_code;
 282         vm_fault_t fault;
 283         unsigned long address, mm_rss;
 284         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 285
 286         fault_code = get_thread_fault_code();
 287
 288         if (notify_page_fault(regs))
 289                 goto exit_exception;
 290
 291         si_code = SEGV_MAPERR;
 292         address = current_thread_info()->fault_address;
 293
 294         if ((fault_code & FAULT_CODE_ITLB) &&
 295             (fault_code & FAULT_CODE_DTLB))
 296                 BUG();
 297
 298         if (test_thread_flag(TIF_32BIT)) {
 299                 if (!(regs->tstate & TSTATE_PRIV)) {
 300                         if (unlikely((regs->tpc >> 32) != 0)) {
 301                                 bogus_32bit_fault_tpc(regs);
 302                                 goto intr_or_no_mm;
 303                         }
 304                 }
 305                 if (unlikely((address >> 32) != 0))
 306                         goto intr_or_no_mm;
 307         }
 308
 309         if (regs->tstate & TSTATE_PRIV) {
 310                 unsigned long tpc = regs->tpc;
 311
 312                 /* Sanity check the PC. */
 313                 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
 314                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 315                         /* Valid, no problems... */
 316                 } else {
 317                         bad_kernel_pc(regs, address);
 318                         goto exit_exception;
 319                 }
 320         } else
 321                 flags |= FAULT_FLAG_USER;
 322
 323         /*
 324          * If we're in an interrupt or have no user
 325          * context, we must not take the fault..
 326          */
 327         if (faulthandler_disabled() || !mm)
 328                 goto intr_or_no_mm;
 329
 330         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 331
 332         if (!down_read_trylock(&mm->mmap_sem)) {
 333                 if ((regs->tstate & TSTATE_PRIV) &&
 334                     !search_exception_tables(regs->tpc)) {
 335                         insn = get_fault_insn(regs, insn);
 336                         goto handle_kernel_fault;
 337                 }
 338
 339 retry:
 340                 down_read(&mm->mmap_sem);
 341         }
 342
 343         if (fault_code & FAULT_CODE_BAD_RA)
 344                 goto do_sigbus;
 345
 346         vma = find_vma(mm, address);
 347         if (!vma)
 348                 goto bad_area;
 349
 350         /* Pure DTLB misses do not tell us whether the fault causing
 351          * load/store/atomic was a write or not, it only says that there
 352          * was no match.  So in such a case we (carefully) read the
 353          * instruction to try and figure this out.  It's an optimization
 354          * so it's ok if we can't do this.
 355          *
 356          * Special hack, window spill/fill knows the exact fault type.
 357          */
 358         if (((fault_code &
 359               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 360             (vma->vm_flags & VM_WRITE) != 0) {
 361                 insn = get_fault_insn(regs, 0);
 362                 if (!insn)
 363                         goto continue_fault;
 364                 /* All loads, stores and atomics have bits 30 and 31 both set
 365                  * in the instruction.  Bit 21 is set in all stores, but we
 366                  * have to avoid prefetches which also have bit 21 set.
 367                  */
 368                 if ((insn & 0xc0200000) == 0xc0200000 &&
 369                     (insn & 0x01780000) != 0x01680000) {
 370                         /* Don't bother updating thread struct value,
 371                          * because update_mmu_cache only cares which tlb
 372                          * the access came from.
 373                          */
 374                         fault_code |= FAULT_CODE_WRITE;
 375                 }
 376         }
 377 continue_fault:
 378
 379         if (vma->vm_start <= address)
 380                 goto good_area;
 381         if (!(vma->vm_flags & VM_GROWSDOWN))
 382                 goto bad_area;
 383         if (!(fault_code & FAULT_CODE_WRITE)) {
 384                 /* Non-faulting loads shouldn't expand stack. */
 385                 insn = get_fault_insn(regs, insn);
 386                 if ((insn & 0xc0800000) == 0xc0800000) {
 387                         unsigned char asi;
 388
 389                         if (insn & 0x2000)
 390                                 asi = (regs->tstate >> 24);
 391                         else
 392                                 asi = (insn >> 5);
 393                         if ((asi & 0xf2) == 0x82)
 394                                 goto bad_area;
 395                 }
 396         }
 397         if (expand_stack(vma, address))
 398                 goto bad_area;
 399         /*
 400          * Ok, we have a good vm_area for this memory access, so
 401          * we can handle it..
 402          */
 403 good_area:
 404         si_code = SEGV_ACCERR;
 405
 406         /* If we took a ITLB miss on a non-executable page, catch
 407          * that here.
 408          */
 409         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 410                 WARN(address != regs->tpc,
 411                      "address (%lx) != regs->tpc (%lx)\n", address, regs->tpc);
 412                 WARN_ON(regs->tstate & TSTATE_PRIV);
 413                 goto bad_area;
 414         }
 415
 416         if (fault_code & FAULT_CODE_WRITE) {
 417                 if (!(vma->vm_flags & VM_WRITE))
 418                         goto bad_area;
 419
 420                 /* Spitfire has an icache which does not snoop
 421                  * processor stores.  Later processors do...
 422                  */
 423                 if (tlb_type == spitfire &&
 424                     (vma->vm_flags & VM_EXEC) != 0 &&
 425                     vma->vm_file != NULL)
 426                         set_thread_fault_code(fault_code |
 427                                               FAULT_CODE_BLKCOMMIT);
 428
 429                 flags |= FAULT_FLAG_WRITE;
 430         } else {
 431                 /* Allow reads even for write-only mappings */
 432                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 433                         goto bad_area;
 434         }
 435
 436         fault = handle_mm_fault(vma, address, flags);
 437
 438         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 439                 goto exit_exception;
 440
 441         if (unlikely(fault & VM_FAULT_ERROR)) {
 442                 if (fault & VM_FAULT_OOM)
 443                         goto out_of_memory;
 444                 else if (fault & VM_FAULT_SIGSEGV)
 445                         goto bad_area;
 446                 else if (fault & VM_FAULT_SIGBUS)
 447                         goto do_sigbus;
 448                 BUG();
 449         }
 450
 451         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 452                 if (fault & VM_FAULT_MAJOR) {
 453                         current->maj_flt++;
 454                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
 455                                       1, regs, address);
 456                 } else {
 457                         current->min_flt++;
 458                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
 459                                       1, regs, address);
 460                 }
 461                 if (fault & VM_FAULT_RETRY) {
 462                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 463                         flags |= FAULT_FLAG_TRIED;
 464
 465                         /* No need to up_read(&mm->mmap_sem) as we would
 466                          * have already released it in __lock_page_or_retry
 467                          * in mm/filemap.c.
 468                          */
 469
 470                         goto retry;
 471                 }
 472         }
 473         up_read(&mm->mmap_sem);
 474
 475         mm_rss = get_mm_rss(mm);
 476 #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
 477         mm_rss -= (mm->context.thp_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 478 #endif
 479         if (unlikely(mm_rss >
 480                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 481                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 482 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
 483         mm_rss = mm->context.hugetlb_pte_count + mm->context.thp_pte_count;
 484         mm_rss *= REAL_HPAGE_PER_HPAGE;
 485         if (unlikely(mm_rss >
 486                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
 487                 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
 488                         tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 489                 else
 490                         hugetlb_setup(regs);
 491
 492         }
 493 #endif
 494 exit_exception:
 495         exception_exit(prev_state);
 496         return;
 497
 498         /*
 499          * Something tried to access memory that isn't in our memory map..
 500          * Fix it, but check if it's kernel or user first..
 501          */
 502 bad_area:
 503         insn = get_fault_insn(regs, insn);
 504         up_read(&mm->mmap_sem);
 505
 506 handle_kernel_fault:
 507         do_kernel_fault(regs, si_code, fault_code, insn, address);
 508         goto exit_exception;
 509
 510 /*
 511  * We ran out of memory, or some other thing happened to us that made
 512  * us unable to handle the page fault gracefully.
 513  */
 514 out_of_memory:
 515         insn = get_fault_insn(regs, insn);
 516         up_read(&mm->mmap_sem);
 517         if (!(regs->tstate & TSTATE_PRIV)) {
 518                 pagefault_out_of_memory();
 519                 goto exit_exception;
 520         }
 521         goto handle_kernel_fault;
 522
 523 intr_or_no_mm:
 524         insn = get_fault_insn(regs, 0);
 525         goto handle_kernel_fault;
 526
 527 do_sigbus:
 528         insn = get_fault_insn(regs, insn);
 529         up_read(&mm->mmap_sem);
 530
 531         /*
 532          * Send a sigbus, regardless of whether we were in kernel
 533          * or user mode.
 534          */
 535         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
 536
 537         /* Kernel mode? Handle exceptions or die */
 538         if (regs->tstate & TSTATE_PRIV)
 539                 goto handle_kernel_fault;
 540 }