arch/cris/mm/fault.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  arch/cris/mm/fault.c
   4  *
   5  *  Copyright (C) 2000-2010  Axis Communications AB
   6  */
   7
   8 #include <linux/mm.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/extable.h>
  11 #include <linux/wait.h>
  12 #include <linux/sched/signal.h>
  13 #include <linux/uaccess.h>
  14 #include <arch/system.h>
  15
  16 extern int find_fixup_code(struct pt_regs *);
  17 extern void die_if_kernel(const char *, struct pt_regs *, long);
  18 extern void show_registers(struct pt_regs *regs);
  19
  20 /* debug of low-level TLB reload */
  21 #undef DEBUG
  22
  23 #ifdef DEBUG
  24 #define D(x) x
  25 #else
  26 #define D(x)
  27 #endif
  28
  29 /* debug of higher-level faults */
  30 #define DPG(x)
  31
  32 /* current active page directory */
  33
  34 DEFINE_PER_CPU(pgd_t *, current_pgd);
  35 unsigned long cris_signal_return_page;
  36
  37 /*
  38  * This routine handles page faults.  It determines the address,
  39  * and the problem, and then passes it off to one of the appropriate
  40  * routines.
  41  *
  42  * Notice that the address we're given is aligned to the page the fault
  43  * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
  44  * address.
  45  *
  46  * error_code:
  47  *      bit 0 == 0 means no page found, 1 means protection fault
  48  *      bit 1 == 0 means read, 1 means write
  49  *
  50  * If this routine detects a bad access, it returns 1, otherwise it
  51  * returns 0.
  52  */
  53
  54 asmlinkage void
  55 do_page_fault(unsigned long address, struct pt_regs *regs,
  56               int protection, int writeaccess)
  57 {
  58         struct task_struct *tsk;
  59         struct mm_struct *mm;
  60         struct vm_area_struct * vma;
  61         siginfo_t info;
  62         int fault;
  63         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  64
  65         D(printk(KERN_DEBUG
  66                  "Page fault for %lX on %X at %lX, prot %d write %d\n",
  67                  address, smp_processor_id(), instruction_pointer(regs),
  68                  protection, writeaccess));
  69
  70         tsk = current;
  71
  72         /*
  73          * We fault-in kernel-space virtual memory on-demand. The
  74          * 'reference' page table is init_mm.pgd.
  75          *
  76          * NOTE! We MUST NOT take any locks for this case. We may
  77          * be in an interrupt or a critical region, and should
  78          * only copy the information from the master page table,
  79          * nothing more.
  80          *
  81          * NOTE2: This is done so that, when updating the vmalloc
  82          * mappings we don't have to walk all processes pgdirs and
  83          * add the high mappings all at once. Instead we do it as they
  84          * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
  85          * bit set so sometimes the TLB can use a lingering entry.
  86          *
  87          * This verifies that the fault happens in kernel space
  88          * and that the fault was not a protection error (error_code & 1).
  89          */
  90
  91         if (address >= VMALLOC_START &&
  92             !protection &&
  93             !user_mode(regs))
  94                 goto vmalloc_fault;
  95
  96         /* When stack execution is not allowed we store the signal
  97          * trampolines in the reserved cris_signal_return_page.
  98          * Handle this in the exact same way as vmalloc (we know
  99          * that the mapping is there and is valid so no need to
 100          * call handle_mm_fault).
 101          */
 102         if (cris_signal_return_page &&
 103             address == cris_signal_return_page &&
 104             !protection && user_mode(regs))
 105                 goto vmalloc_fault;
 106
 107         /* we can and should enable interrupts at this point */
 108         local_irq_enable();
 109
 110         mm = tsk->mm;
 111         info.si_code = SEGV_MAPERR;
 112
 113         /*
 114          * If we're in an interrupt, have pagefaults disabled or have no
 115          * user context, we must not take the fault.
 116          */
 117
 118         if (faulthandler_disabled() || !mm)
 119                 goto no_context;
 120
 121         if (user_mode(regs))
 122                 flags |= FAULT_FLAG_USER;
 123 retry:
 124         down_read(&mm->mmap_sem);
 125         vma = find_vma(mm, address);
 126         if (!vma)
 127                 goto bad_area;
 128         if (vma->vm_start <= address)
 129                 goto good_area;
 130         if (!(vma->vm_flags & VM_GROWSDOWN))
 131                 goto bad_area;
 132         if (user_mode(regs)) {
 133                 /*
 134                  * accessing the stack below usp is always a bug.
 135                  * we get page-aligned addresses so we can only check
 136                  * if we're within a page from usp, but that might be
 137                  * enough to catch brutal errors at least.
 138                  */
 139                 if (address + PAGE_SIZE < rdusp())
 140                         goto bad_area;
 141         }
 142         if (expand_stack(vma, address))
 143                 goto bad_area;
 144
 145         /*
 146          * Ok, we have a good vm_area for this memory access, so
 147          * we can handle it..
 148          */
 149
 150  good_area:
 151         info.si_code = SEGV_ACCERR;
 152
 153         /* first do some preliminary protection checks */
 154
 155         if (writeaccess == 2){
 156                 if (!(vma->vm_flags & VM_EXEC))
 157                         goto bad_area;
 158         } else if (writeaccess == 1) {
 159                 if (!(vma->vm_flags & VM_WRITE))
 160                         goto bad_area;
 161                 flags |= FAULT_FLAG_WRITE;
 162         } else {
 163                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 164                         goto bad_area;
 165         }
 166
 167         /*
 168          * If for any reason at all we couldn't handle the fault,
 169          * make sure we exit gracefully rather than endlessly redo
 170          * the fault.
 171          */
 172
 173         fault = handle_mm_fault(vma, address, flags);
 174
 175         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 176                 return;
 177
 178         if (unlikely(fault & VM_FAULT_ERROR)) {
 179                 if (fault & VM_FAULT_OOM)
 180                         goto out_of_memory;
 181                 else if (fault & VM_FAULT_SIGSEGV)
 182                         goto bad_area;
 183                 else if (fault & VM_FAULT_SIGBUS)
 184                         goto do_sigbus;
 185                 BUG();
 186         }
 187
 188         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 189                 if (fault & VM_FAULT_MAJOR)
 190                         tsk->maj_flt++;
 191                 else
 192                         tsk->min_flt++;
 193                 if (fault & VM_FAULT_RETRY) {
 194                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 195                         flags |= FAULT_FLAG_TRIED;
 196
 197                         /*
 198                          * No need to up_read(&mm->mmap_sem) as we would
 199                          * have already released it in __lock_page_or_retry
 200                          * in mm/filemap.c.
 201                          */
 202
 203                         goto retry;
 204                 }
 205         }
 206
 207         up_read(&mm->mmap_sem);
 208         return;
 209
 210         /*
 211          * Something tried to access memory that isn't in our memory map..
 212          * Fix it, but check if it's kernel or user first..
 213          */
 214
 215  bad_area:
 216         up_read(&mm->mmap_sem);
 217
 218  bad_area_nosemaphore:
 219         DPG(show_registers(regs));
 220
 221         /* User mode accesses just cause a SIGSEGV */
 222
 223         if (user_mode(regs)) {
 224 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
 225                 DECLARE_WAIT_QUEUE_HEAD(wq);
 226 #endif
 227                 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
 228                         "address %08lx at pc %08lx\n",
 229                         tsk->comm, tsk->pid,
 230                         address, instruction_pointer(regs));
 231
 232                 /* With DPG on, we've already dumped registers above.  */
 233                 DPG(if (0))
 234                         show_registers(regs);
 235
 236 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
 237                 wait_event_interruptible(wq, 0 == 1);
 238 #else
 239                 info.si_signo = SIGSEGV;
 240                 info.si_errno = 0;
 241                 /* info.si_code has been set above */
 242                 info.si_addr = (void *)address;
 243                 force_sig_info(SIGSEGV, &info, tsk);
 244 #endif
 245                 return;
 246         }
 247
 248  no_context:
 249
 250         /* Are we prepared to handle this kernel fault?
 251          *
 252          * (The kernel has valid exception-points in the source
 253          *  when it accesses user-memory. When it fails in one
 254          *  of those points, we find it in a table and do a jump
 255          *  to some fixup code that loads an appropriate error
 256          *  code)
 257          */
 258
 259         if (find_fixup_code(regs))
 260                 return;
 261
 262         /*
 263          * Oops. The kernel tried to access some bad page. We'll have to
 264          * terminate things with extreme prejudice.
 265          */
 266
 267         if (!oops_in_progress) {
 268                 oops_in_progress = 1;
 269                 if ((unsigned long) (address) < PAGE_SIZE)
 270                         printk(KERN_ALERT "Unable to handle kernel NULL "
 271                                 "pointer dereference");
 272                 else
 273                         printk(KERN_ALERT "Unable to handle kernel access"
 274                                 " at virtual address %08lx\n", address);
 275
 276                 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
 277                 oops_in_progress = 0;
 278         }
 279
 280         do_exit(SIGKILL);
 281
 282         /*
 283          * We ran out of memory, or some other thing happened to us that made
 284          * us unable to handle the page fault gracefully.
 285          */
 286
 287  out_of_memory:
 288         up_read(&mm->mmap_sem);
 289         if (!user_mode(regs))
 290                 goto no_context;
 291         pagefault_out_of_memory();
 292         return;
 293
 294  do_sigbus:
 295         up_read(&mm->mmap_sem);
 296
 297         /*
 298          * Send a sigbus, regardless of whether we were in kernel
 299          * or user mode.
 300          */
 301         info.si_signo = SIGBUS;
 302         info.si_errno = 0;
 303         info.si_code = BUS_ADRERR;
 304         info.si_addr = (void *)address;
 305         force_sig_info(SIGBUS, &info, tsk);
 306
 307         /* Kernel mode? Handle exceptions or die */
 308         if (!user_mode(regs))
 309                 goto no_context;
 310         return;
 311
 312 vmalloc_fault:
 313         {
 314                 /*
 315                  * Synchronize this task's top level page-table
 316                  * with the 'reference' page table.
 317                  *
 318                  * Use current_pgd instead of tsk->active_mm->pgd
 319                  * since the latter might be unavailable if this
 320                  * code is executed in a misfortunately run irq
 321                  * (like inside schedule() between switch_mm and
 322                  *  switch_to...).
 323                  */
 324
 325                 int offset = pgd_index(address);
 326                 pgd_t *pgd, *pgd_k;
 327                 pud_t *pud, *pud_k;
 328                 pmd_t *pmd, *pmd_k;
 329                 pte_t *pte_k;
 330
 331                 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
 332                 pgd_k = init_mm.pgd + offset;
 333
 334                 /* Since we're two-level, we don't need to do both
 335                  * set_pgd and set_pmd (they do the same thing). If
 336                  * we go three-level at some point, do the right thing
 337                  * with pgd_present and set_pgd here.
 338                  *
 339                  * Also, since the vmalloc area is global, we don't
 340                  * need to copy individual PTE's, it is enough to
 341                  * copy the pgd pointer into the pte page of the
 342                  * root task. If that is there, we'll find our pte if
 343                  * it exists.
 344                  */
 345
 346                 pud = pud_offset(pgd, address);
 347                 pud_k = pud_offset(pgd_k, address);
 348                 if (!pud_present(*pud_k))
 349                         goto no_context;
 350
 351                 pmd = pmd_offset(pud, address);
 352                 pmd_k = pmd_offset(pud_k, address);
 353
 354                 if (!pmd_present(*pmd_k))
 355                         goto bad_area_nosemaphore;
 356
 357                 set_pmd(pmd, *pmd_k);
 358
 359                 /* Make sure the actual PTE exists as well to
 360                  * catch kernel vmalloc-area accesses to non-mapped
 361                  * addresses. If we don't do this, this will just
 362                  * silently loop forever.
 363                  */
 364
 365                 pte_k = pte_offset_kernel(pmd_k, address);
 366                 if (!pte_present(*pte_k))
 367                         goto no_context;
 368
 369                 return;
 370         }
 371 }
 372
 373 /* Find fixup code. */
 374 int
 375 find_fixup_code(struct pt_regs *regs)
 376 {
 377         const struct exception_table_entry *fixup;
 378         /* in case of delay slot fault (v32) */
 379         unsigned long ip = (instruction_pointer(regs) & ~0x1);
 380
 381         fixup = search_exception_tables(ip);
 382         if (fixup != 0) {
 383                 /* Adjust the instruction pointer in the stackframe. */
 384                 instruction_pointer(regs) = fixup->fixup;
 385                 arch_fixup(regs);
 386                 return 1;
 387         }
 388
 389         return 0;
 390 }