arch/nds32/mm/fault.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 // Copyright (C) 2005-2017 Andes Technology Corporation
   3
   4 #include <linux/extable.h>
   5 #include <linux/module.h>
   6 #include <linux/signal.h>
   7 #include <linux/ptrace.h>
   8 #include <linux/mm.h>
   9 #include <linux/init.h>
  10 #include <linux/hardirq.h>
  11 #include <linux/uaccess.h>
  12
  13 #include <asm/pgtable.h>
  14 #include <asm/tlbflush.h>
  15
  16 extern void die(const char *str, struct pt_regs *regs, long err);
  17
  18 /*
  19  * This is useful to dump out the page tables associated with
  20  * 'addr' in mm 'mm'.
  21  */
  22 void show_pte(struct mm_struct *mm, unsigned long addr)
  23 {
  24         pgd_t *pgd;
  25         if (!mm)
  26                 mm = &init_mm;
  27
  28         pr_alert("pgd = %p\n", mm->pgd);
  29         pgd = pgd_offset(mm, addr);
  30         pr_alert("[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
  31
  32         do {
  33                 pmd_t *pmd;
  34
  35                 if (pgd_none(*pgd))
  36                         break;
  37
  38                 if (pgd_bad(*pgd)) {
  39                         pr_alert("(bad)");
  40                         break;
  41                 }
  42
  43                 pmd = pmd_offset(pgd, addr);
  44 #if PTRS_PER_PMD != 1
  45                 pr_alert(", *pmd=%08lx", pmd_val(*pmd));
  46 #endif
  47
  48                 if (pmd_none(*pmd))
  49                         break;
  50
  51                 if (pmd_bad(*pmd)) {
  52                         pr_alert("(bad)");
  53                         break;
  54                 }
  55
  56                 if (IS_ENABLED(CONFIG_HIGHMEM))
  57                 {
  58                         pte_t *pte;
  59                         /* We must not map this if we have highmem enabled */
  60                         pte = pte_offset_map(pmd, addr);
  61                         pr_alert(", *pte=%08lx", pte_val(*pte));
  62                         pte_unmap(pte);
  63                 }
  64         } while (0);
  65
  66         pr_alert("\n");
  67 }
  68
  69 void do_page_fault(unsigned long entry, unsigned long addr,
  70                    unsigned int error_code, struct pt_regs *regs)
  71 {
  72         struct task_struct *tsk;
  73         struct mm_struct *mm;
  74         struct vm_area_struct *vma;
  75         siginfo_t info;
  76         int fault;
  77         unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
  78         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  79
  80         clear_siginfo(&info);
  81         error_code = error_code & (ITYPE_mskINST | ITYPE_mskETYPE);
  82         tsk = current;
  83         mm = tsk->mm;
  84         info.si_code = SEGV_MAPERR;
  85         /*
  86          * We fault-in kernel-space virtual memory on-demand. The
  87          * 'reference' page table is init_mm.pgd.
  88          *
  89          * NOTE! We MUST NOT take any locks for this case. We may
  90          * be in an interrupt or a critical region, and should
  91          * only copy the information from the master page table,
  92          * nothing more.
  93          */
  94         if (addr >= TASK_SIZE) {
  95                 if (user_mode(regs))
  96                         goto bad_area_nosemaphore;
  97
  98                 if (addr >= TASK_SIZE && addr < VMALLOC_END
  99                     && (entry == ENTRY_PTE_NOT_PRESENT))
 100                         goto vmalloc_fault;
 101                 else
 102                         goto no_context;
 103         }
 104
 105         /* Send a signal to the task for handling the unalignment access. */
 106         if (entry == ENTRY_GENERAL_EXCPETION
 107             && error_code == ETYPE_ALIGNMENT_CHECK) {
 108                 if (user_mode(regs))
 109                         goto bad_area_nosemaphore;
 110                 else
 111                         goto no_context;
 112         }
 113
 114         /*
 115          * If we're in an interrupt or have no user
 116          * context, we must not take the fault..
 117          */
 118         if (unlikely(faulthandler_disabled() || !mm))
 119                 goto no_context;
 120
 121         /*
 122          * As per x86, we may deadlock here. However, since the kernel only
 123          * validly references user space from well defined areas of the code,
 124          * we can bug out early if this is from code which shouldn't.
 125          */
 126         if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
 127                 if (!user_mode(regs) &&
 128                     !search_exception_tables(instruction_pointer(regs)))
 129                         goto no_context;
 130 retry:
 131                 down_read(&mm->mmap_sem);
 132         } else {
 133                 /*
 134                  * The above down_read_trylock() might have succeeded in which
 135                  * case, we'll have missed the might_sleep() from down_read().
 136                  */
 137                 might_sleep();
 138                 if (IS_ENABLED(CONFIG_DEBUG_VM)) {
 139                         if (!user_mode(regs) &&
 140                             !search_exception_tables(instruction_pointer(regs)))
 141                                 goto no_context;
 142                 }
 143         }
 144
 145         vma = find_vma(mm, addr);
 146
 147         if (unlikely(!vma))
 148                 goto bad_area;
 149
 150         if (vma->vm_start <= addr)
 151                 goto good_area;
 152
 153         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
 154                 goto bad_area;
 155
 156         if (unlikely(expand_stack(vma, addr)))
 157                 goto bad_area;
 158
 159         /*
 160          * Ok, we have a good vm_area for this memory access, so
 161          * we can handle it..
 162          */
 163
 164 good_area:
 165         info.si_code = SEGV_ACCERR;
 166
 167         /* first do some preliminary protection checks */
 168         if (entry == ENTRY_PTE_NOT_PRESENT) {
 169                 if (error_code & ITYPE_mskINST)
 170                         mask = VM_EXEC;
 171                 else {
 172                         mask = VM_READ | VM_WRITE;
 173                         if (vma->vm_flags & VM_WRITE)
 174                                 flags |= FAULT_FLAG_WRITE;
 175                 }
 176         } else if (entry == ENTRY_TLB_MISC) {
 177                 switch (error_code & ITYPE_mskETYPE) {
 178                 case RD_PROT:
 179                         mask = VM_READ;
 180                         break;
 181                 case WRT_PROT:
 182                         mask = VM_WRITE;
 183                         flags |= FAULT_FLAG_WRITE;
 184                         break;
 185                 case NOEXEC:
 186                         mask = VM_EXEC;
 187                         break;
 188                 case PAGE_MODIFY:
 189                         mask = VM_WRITE;
 190                         flags |= FAULT_FLAG_WRITE;
 191                         break;
 192                 case ACC_BIT:
 193                         BUG();
 194                 default:
 195                         break;
 196                 }
 197
 198         }
 199         if (!(vma->vm_flags & mask))
 200                 goto bad_area;
 201
 202         /*
 203          * If for any reason at all we couldn't handle the fault,
 204          * make sure we exit gracefully rather than endlessly redo
 205          * the fault.
 206          */
 207
 208         fault = handle_mm_fault(vma, addr, flags);
 209
 210         /*
 211          * If we need to retry but a fatal signal is pending, handle the
 212          * signal first. We do not need to release the mmap_sem because it
 213          * would already be released in __lock_page_or_retry in mm/filemap.c.
 214          */
 215         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
 216                 if (!user_mode(regs))
 217                         goto no_context;
 218                 return;
 219         }
 220
 221         if (unlikely(fault & VM_FAULT_ERROR)) {
 222                 if (fault & VM_FAULT_OOM)
 223                         goto out_of_memory;
 224                 else if (fault & VM_FAULT_SIGBUS)
 225                         goto do_sigbus;
 226                 else
 227                         goto bad_area;
 228         }
 229
 230         /*
 231          * Major/minor page fault accounting is only done on the initial
 232          * attempt. If we go through a retry, it is extremely likely that the
 233          * page will be found in page cache at that point.
 234          */
 235         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 236                 if (fault & VM_FAULT_MAJOR)
 237                         tsk->maj_flt++;
 238                 else
 239                         tsk->min_flt++;
 240                 if (fault & VM_FAULT_RETRY) {
 241                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 242                         flags |= FAULT_FLAG_TRIED;
 243
 244                         /* No need to up_read(&mm->mmap_sem) as we would
 245                          * have already released it in __lock_page_or_retry
 246                          * in mm/filemap.c.
 247                          */
 248                         goto retry;
 249                 }
 250         }
 251
 252         up_read(&mm->mmap_sem);
 253         return;
 254
 255         /*
 256          * Something tried to access memory that isn't in our memory map..
 257          * Fix it, but check if it's kernel or user first..
 258          */
 259 bad_area:
 260         up_read(&mm->mmap_sem);
 261
 262 bad_area_nosemaphore:
 263
 264         /* User mode accesses just cause a SIGSEGV */
 265
 266         if (user_mode(regs)) {
 267                 tsk->thread.address = addr;
 268                 tsk->thread.error_code = error_code;
 269                 tsk->thread.trap_no = entry;
 270                 info.si_signo = SIGSEGV;
 271                 info.si_errno = 0;
 272                 /* info.si_code has been set above */
 273                 info.si_addr = (void *)addr;
 274                 force_sig_info(SIGSEGV, &info, tsk);
 275                 return;
 276         }
 277
 278 no_context:
 279
 280         /* Are we prepared to handle this kernel fault?
 281          *
 282          * (The kernel has valid exception-points in the source
 283          *  when it acesses user-memory. When it fails in one
 284          *  of those points, we find it in a table and do a jump
 285          *  to some fixup code that loads an appropriate error
 286          *  code)
 287          */
 288
 289         {
 290                 const struct exception_table_entry *entry;
 291
 292                 if ((entry =
 293                      search_exception_tables(instruction_pointer(regs))) !=
 294                     NULL) {
 295                         /* Adjust the instruction pointer in the stackframe */
 296                         instruction_pointer(regs) = entry->fixup;
 297                         return;
 298                 }
 299         }
 300
 301         /*
 302          * Oops. The kernel tried to access some bad page. We'll have to
 303          * terminate things with extreme prejudice.
 304          */
 305
 306         bust_spinlocks(1);
 307         pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
 308                  (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 309                  "paging request", addr);
 310
 311         show_pte(mm, addr);
 312         die("Oops", regs, error_code);
 313         bust_spinlocks(0);
 314         do_exit(SIGKILL);
 315
 316         return;
 317
 318         /*
 319          * We ran out of memory, or some other thing happened to us that made
 320          * us unable to handle the page fault gracefully.
 321          */
 322
 323 out_of_memory:
 324         up_read(&mm->mmap_sem);
 325         if (!user_mode(regs))
 326                 goto no_context;
 327         pagefault_out_of_memory();
 328         return;
 329
 330 do_sigbus:
 331         up_read(&mm->mmap_sem);
 332
 333         /* Kernel mode? Handle exceptions or die */
 334         if (!user_mode(regs))
 335                 goto no_context;
 336
 337         /*
 338          * Send a sigbus
 339          */
 340         tsk->thread.address = addr;
 341         tsk->thread.error_code = error_code;
 342         tsk->thread.trap_no = entry;
 343         info.si_signo = SIGBUS;
 344         info.si_errno = 0;
 345         info.si_code = BUS_ADRERR;
 346         info.si_addr = (void *)addr;
 347         force_sig_info(SIGBUS, &info, tsk);
 348
 349         return;
 350
 351 vmalloc_fault:
 352         {
 353                 /*
 354                  * Synchronize this task's top level page-table
 355                  * with the 'reference' page table.
 356                  *
 357                  * Use current_pgd instead of tsk->active_mm->pgd
 358                  * since the latter might be unavailable if this
 359                  * code is executed in a misfortunately run irq
 360                  * (like inside schedule() between switch_mm and
 361                  *  switch_to...).
 362                  */
 363
 364                 unsigned int index = pgd_index(addr);
 365                 pgd_t *pgd, *pgd_k;
 366                 pud_t *pud, *pud_k;
 367                 pmd_t *pmd, *pmd_k;
 368                 pte_t *pte_k;
 369
 370                 pgd = (pgd_t *) __va(__nds32__mfsr(NDS32_SR_L1_PPTB)) + index;
 371                 pgd_k = init_mm.pgd + index;
 372
 373                 if (!pgd_present(*pgd_k))
 374                         goto no_context;
 375
 376                 pud = pud_offset(pgd, addr);
 377                 pud_k = pud_offset(pgd_k, addr);
 378                 if (!pud_present(*pud_k))
 379                         goto no_context;
 380
 381                 pmd = pmd_offset(pud, addr);
 382                 pmd_k = pmd_offset(pud_k, addr);
 383                 if (!pmd_present(*pmd_k))
 384                         goto no_context;
 385
 386                 if (!pmd_present(*pmd))
 387                         set_pmd(pmd, *pmd_k);
 388                 else
 389                         BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
 390
 391                 /*
 392                  * Since the vmalloc area is global, we don't
 393                  * need to copy individual PTE's, it is enough to
 394                  * copy the pgd pointer into the pte page of the
 395                  * root task. If that is there, we'll find our pte if
 396                  * it exists.
 397                  */
 398
 399                 /* Make sure the actual PTE exists as well to
 400                  * catch kernel vmalloc-area accesses to non-mapped
 401                  * addres. If we don't do this, this will just
 402                  * silently loop forever.
 403                  */
 404
 405                 pte_k = pte_offset_kernel(pmd_k, addr);
 406                 if (!pte_present(*pte_k))
 407                         goto no_context;
 408
 409                 return;
 410         }
 411 }