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 #include <linux/perf_event.h>
  13
  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                 p4d_t *p4d;
  34                 pud_t *pud;
  35                 pmd_t *pmd;
  36
  37                 if (pgd_none(*pgd))
  38                         break;
  39
  40                 if (pgd_bad(*pgd)) {
  41                         pr_alert("(bad)");
  42                         break;
  43                 }
  44
  45                 p4d = p4d_offset(pgd, addr);
  46                 pud = pud_offset(p4d, addr);
  47                 pmd = pmd_offset(pud, addr);
  48 #if PTRS_PER_PMD != 1
  49                 pr_alert(", *pmd=%08lx", pmd_val(*pmd));
  50 #endif
  51
  52                 if (pmd_none(*pmd))
  53                         break;
  54
  55                 if (pmd_bad(*pmd)) {
  56                         pr_alert("(bad)");
  57                         break;
  58                 }
  59
  60                 if (IS_ENABLED(CONFIG_HIGHMEM))
  61                 {
  62                         pte_t *pte;
  63                         /* We must not map this if we have highmem enabled */
  64                         pte = pte_offset_map(pmd, addr);
  65                         pr_alert(", *pte=%08lx", pte_val(*pte));
  66                         pte_unmap(pte);
  67                 }
  68         } while (0);
  69
  70         pr_alert("\n");
  71 }
  72
  73 void do_page_fault(unsigned long entry, unsigned long addr,
  74                    unsigned int error_code, struct pt_regs *regs)
  75 {
  76         struct task_struct *tsk;
  77         struct mm_struct *mm;
  78         struct vm_area_struct *vma;
  79         int si_code;
  80         vm_fault_t fault;
  81         unsigned int mask = VM_ACCESS_FLAGS;
  82         unsigned int flags = FAULT_FLAG_DEFAULT;
  83
  84         error_code = error_code & (ITYPE_mskINST | ITYPE_mskETYPE);
  85         tsk = current;
  86         mm = tsk->mm;
  87         si_code = SEGV_MAPERR;
  88         /*
  89          * We fault-in kernel-space virtual memory on-demand. The
  90          * 'reference' page table is init_mm.pgd.
  91          *
  92          * NOTE! We MUST NOT take any locks for this case. We may
  93          * be in an interrupt or a critical region, and should
  94          * only copy the information from the master page table,
  95          * nothing more.
  96          */
  97         if (addr >= TASK_SIZE) {
  98                 if (user_mode(regs))
  99                         goto bad_area_nosemaphore;
 100
 101                 if (addr >= TASK_SIZE && addr < VMALLOC_END
 102                     && (entry == ENTRY_PTE_NOT_PRESENT))
 103                         goto vmalloc_fault;
 104                 else
 105                         goto no_context;
 106         }
 107
 108         /* Send a signal to the task for handling the unalignment access. */
 109         if (entry == ENTRY_GENERAL_EXCPETION
 110             && error_code == ETYPE_ALIGNMENT_CHECK) {
 111                 if (user_mode(regs))
 112                         goto bad_area_nosemaphore;
 113                 else
 114                         goto no_context;
 115         }
 116
 117         /*
 118          * If we're in an interrupt or have no user
 119          * context, we must not take the fault..
 120          */
 121         if (unlikely(faulthandler_disabled() || !mm))
 122                 goto no_context;
 123
 124         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
 125
 126         /*
 127          * As per x86, we may deadlock here. However, since the kernel only
 128          * validly references user space from well defined areas of the code,
 129          * we can bug out early if this is from code which shouldn't.
 130          */
 131         if (unlikely(!mmap_read_trylock(mm))) {
 132                 if (!user_mode(regs) &&
 133                     !search_exception_tables(instruction_pointer(regs)))
 134                         goto no_context;
 135 retry:
 136                 mmap_read_lock(mm);
 137         } else {
 138                 /*
 139                  * The above down_read_trylock() might have succeeded in which
 140                  * case, we'll have missed the might_sleep() from down_read().
 141                  */
 142                 might_sleep();
 143                 if (IS_ENABLED(CONFIG_DEBUG_VM)) {
 144                         if (!user_mode(regs) &&
 145                             !search_exception_tables(instruction_pointer(regs)))
 146                                 goto no_context;
 147                 }
 148         }
 149
 150         vma = find_vma(mm, addr);
 151
 152         if (unlikely(!vma))
 153                 goto bad_area;
 154
 155         if (vma->vm_start <= addr)
 156                 goto good_area;
 157
 158         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
 159                 goto bad_area;
 160
 161         if (unlikely(expand_stack(vma, addr)))
 162                 goto bad_area;
 163
 164         /*
 165          * Ok, we have a good vm_area for this memory access, so
 166          * we can handle it..
 167          */
 168
 169 good_area:
 170         si_code = SEGV_ACCERR;
 171
 172         /* first do some preliminary protection checks */
 173         if (entry == ENTRY_PTE_NOT_PRESENT) {
 174                 if (error_code & ITYPE_mskINST)
 175                         mask = VM_EXEC;
 176                 else {
 177                         mask = VM_READ | VM_WRITE;
 178                 }
 179         } else if (entry == ENTRY_TLB_MISC) {
 180                 switch (error_code & ITYPE_mskETYPE) {
 181                 case RD_PROT:
 182                         mask = VM_READ;
 183                         break;
 184                 case WRT_PROT:
 185                         mask = VM_WRITE;
 186                         flags |= FAULT_FLAG_WRITE;
 187                         break;
 188                 case NOEXEC:
 189                         mask = VM_EXEC;
 190                         break;
 191                 case PAGE_MODIFY:
 192                         mask = VM_WRITE;
 193                         flags |= FAULT_FLAG_WRITE;
 194                         break;
 195                 case ACC_BIT:
 196                         BUG();
 197                 default:
 198                         break;
 199                 }
 200
 201         }
 202         if (!(vma->vm_flags & mask))
 203                 goto bad_area;
 204
 205         /*
 206          * If for any reason at all we couldn't handle the fault,
 207          * make sure we exit gracefully rather than endlessly redo
 208          * the fault.
 209          */
 210
 211         fault = handle_mm_fault(vma, addr, flags, regs);
 212
 213         /*
 214          * If we need to retry but a fatal signal is pending, handle the
 215          * signal first. We do not need to release the mmap_lock because it
 216          * would already be released in __lock_page_or_retry in mm/filemap.c.
 217          */
 218         if (fault_signal_pending(fault, regs)) {
 219                 if (!user_mode(regs))
 220                         goto no_context;
 221                 return;
 222         }
 223
 224         if (unlikely(fault & VM_FAULT_ERROR)) {
 225                 if (fault & VM_FAULT_OOM)
 226                         goto out_of_memory;
 227                 else if (fault & VM_FAULT_SIGBUS)
 228                         goto do_sigbus;
 229                 else
 230                         goto bad_area;
 231         }
 232
 233         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 234                 if (fault & VM_FAULT_RETRY) {
 235                         flags |= FAULT_FLAG_TRIED;
 236
 237                         /* No need to mmap_read_unlock(mm) as we would
 238                          * have already released it in __lock_page_or_retry
 239                          * in mm/filemap.c.
 240                          */
 241                         goto retry;
 242                 }
 243         }
 244
 245         mmap_read_unlock(mm);
 246         return;
 247
 248         /*
 249          * Something tried to access memory that isn't in our memory map..
 250          * Fix it, but check if it's kernel or user first..
 251          */
 252 bad_area:
 253         mmap_read_unlock(mm);
 254
 255 bad_area_nosemaphore:
 256
 257         /* User mode accesses just cause a SIGSEGV */
 258
 259         if (user_mode(regs)) {
 260                 tsk->thread.address = addr;
 261                 tsk->thread.error_code = error_code;
 262                 tsk->thread.trap_no = entry;
 263                 force_sig_fault(SIGSEGV, si_code, (void __user *)addr);
 264                 return;
 265         }
 266
 267 no_context:
 268
 269         /* Are we prepared to handle this kernel fault?
 270          *
 271          * (The kernel has valid exception-points in the source
 272          *  when it acesses user-memory. When it fails in one
 273          *  of those points, we find it in a table and do a jump
 274          *  to some fixup code that loads an appropriate error
 275          *  code)
 276          */
 277
 278         {
 279                 const struct exception_table_entry *entry;
 280
 281                 if ((entry =
 282                      search_exception_tables(instruction_pointer(regs))) !=
 283                     NULL) {
 284                         /* Adjust the instruction pointer in the stackframe */
 285                         instruction_pointer(regs) = entry->fixup;
 286                         return;
 287                 }
 288         }
 289
 290         /*
 291          * Oops. The kernel tried to access some bad page. We'll have to
 292          * terminate things with extreme prejudice.
 293          */
 294
 295         bust_spinlocks(1);
 296         pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
 297                  (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 298                  "paging request", addr);
 299
 300         show_pte(mm, addr);
 301         die("Oops", regs, error_code);
 302         bust_spinlocks(0);
 303         do_exit(SIGKILL);
 304
 305         return;
 306
 307         /*
 308          * We ran out of memory, or some other thing happened to us that made
 309          * us unable to handle the page fault gracefully.
 310          */
 311
 312 out_of_memory:
 313         mmap_read_unlock(mm);
 314         if (!user_mode(regs))
 315                 goto no_context;
 316         pagefault_out_of_memory();
 317         return;
 318
 319 do_sigbus:
 320         mmap_read_unlock(mm);
 321
 322         /* Kernel mode? Handle exceptions or die */
 323         if (!user_mode(regs))
 324                 goto no_context;
 325
 326         /*
 327          * Send a sigbus
 328          */
 329         tsk->thread.address = addr;
 330         tsk->thread.error_code = error_code;
 331         tsk->thread.trap_no = entry;
 332         force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr);
 333
 334         return;
 335
 336 vmalloc_fault:
 337         {
 338                 /*
 339                  * Synchronize this task's top level page-table
 340                  * with the 'reference' page table.
 341                  *
 342                  * Use current_pgd instead of tsk->active_mm->pgd
 343                  * since the latter might be unavailable if this
 344                  * code is executed in a misfortunately run irq
 345                  * (like inside schedule() between switch_mm and
 346                  *  switch_to...).
 347                  */
 348
 349                 unsigned int index = pgd_index(addr);
 350                 pgd_t *pgd, *pgd_k;
 351                 p4d_t *p4d, *p4d_k;
 352                 pud_t *pud, *pud_k;
 353                 pmd_t *pmd, *pmd_k;
 354                 pte_t *pte_k;
 355
 356                 pgd = (pgd_t *) __va(__nds32__mfsr(NDS32_SR_L1_PPTB)) + index;
 357                 pgd_k = init_mm.pgd + index;
 358
 359                 if (!pgd_present(*pgd_k))
 360                         goto no_context;
 361
 362                 p4d = p4d_offset(pgd, addr);
 363                 p4d_k = p4d_offset(pgd_k, addr);
 364                 if (!p4d_present(*p4d_k))
 365                         goto no_context;
 366
 367                 pud = pud_offset(p4d, addr);
 368                 pud_k = pud_offset(p4d_k, addr);
 369                 if (!pud_present(*pud_k))
 370                         goto no_context;
 371
 372                 pmd = pmd_offset(pud, addr);
 373                 pmd_k = pmd_offset(pud_k, addr);
 374                 if (!pmd_present(*pmd_k))
 375                         goto no_context;
 376
 377                 if (!pmd_present(*pmd))
 378                         set_pmd(pmd, *pmd_k);
 379                 else
 380                         BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
 381
 382                 /*
 383                  * Since the vmalloc area is global, we don't
 384                  * need to copy individual PTE's, it is enough to
 385                  * copy the pgd pointer into the pte page of the
 386                  * root task. If that is there, we'll find our pte if
 387                  * it exists.
 388                  */
 389
 390                 /* Make sure the actual PTE exists as well to
 391                  * catch kernel vmalloc-area accesses to non-mapped
 392                  * addres. If we don't do this, this will just
 393                  * silently loop forever.
 394                  */
 395
 396                 pte_k = pte_offset_kernel(pmd_k, addr);
 397                 if (!pte_present(*pte_k))
 398                         goto no_context;
 399
 400                 return;
 401         }
 402 }