arch/sh/mm/fault_32.c

   1 /*
   2  * Page fault handler for SH with an MMU.
   3  *
   4  *  Copyright (C) 1999  Niibe Yutaka
   5  *  Copyright (C) 2003 - 2012  Paul Mundt
   6  *
   7  *  Based on linux/arch/i386/mm/fault.c:
   8  *   Copyright (C) 1995  Linus Torvalds
   9  *
  10  * This file is subject to the terms and conditions of the GNU General Public
  11  * License.  See the file "COPYING" in the main directory of this archive
  12  * for more details.
  13  */
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/hardirq.h>
  17 #include <linux/kprobes.h>
  18 #include <linux/perf_event.h>
  19 #include <linux/kdebug.h>
  20 #include <asm/io_trapped.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/tlbflush.h>
  23 #include <asm/traps.h>
  24
  25 static inline int notify_page_fault(struct pt_regs *regs, int trap)
  26 {
  27         int ret = 0;
  28
  29         if (kprobes_built_in() && !user_mode(regs)) {
  30                 preempt_disable();
  31                 if (kprobe_running() && kprobe_fault_handler(regs, trap))
  32                         ret = 1;
  33                 preempt_enable();
  34         }
  35
  36         return ret;
  37 }
  38
  39 static void
  40 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
  41                      struct task_struct *tsk)
  42 {
  43         siginfo_t info;
  44
  45         info.si_signo   = si_signo;
  46         info.si_errno   = 0;
  47         info.si_code    = si_code;
  48         info.si_addr    = (void __user *)address;
  49
  50         force_sig_info(si_signo, &info, tsk);
  51 }
  52
  53 /*
  54  * This is useful to dump out the page tables associated with
  55  * 'addr' in mm 'mm'.
  56  */
  57 static void show_pte(struct mm_struct *mm, unsigned long addr)
  58 {
  59         pgd_t *pgd;
  60
  61         if (mm)
  62                 pgd = mm->pgd;
  63         else
  64                 pgd = get_TTB();
  65
  66         printk(KERN_ALERT "pgd = %p\n", pgd);
  67         pgd += pgd_index(addr);
  68         printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
  69                sizeof(*pgd) * 2, (u64)pgd_val(*pgd));
  70
  71         do {
  72                 pud_t *pud;
  73                 pmd_t *pmd;
  74                 pte_t *pte;
  75
  76                 if (pgd_none(*pgd))
  77                         break;
  78
  79                 if (pgd_bad(*pgd)) {
  80                         printk("(bad)");
  81                         break;
  82                 }
  83
  84                 pud = pud_offset(pgd, addr);
  85                 if (PTRS_PER_PUD != 1)
  86                         printk(", *pud=%0*Lx", sizeof(*pud) * 2,
  87                                (u64)pud_val(*pud));
  88
  89                 if (pud_none(*pud))
  90                         break;
  91
  92                 if (pud_bad(*pud)) {
  93                         printk("(bad)");
  94                         break;
  95                 }
  96
  97                 pmd = pmd_offset(pud, addr);
  98                 if (PTRS_PER_PMD != 1)
  99                         printk(", *pmd=%0*Lx", sizeof(*pmd) * 2,
 100                                (u64)pmd_val(*pmd));
 101
 102                 if (pmd_none(*pmd))
 103                         break;
 104
 105                 if (pmd_bad(*pmd)) {
 106                         printk("(bad)");
 107                         break;
 108                 }
 109
 110                 /* We must not map this if we have highmem enabled */
 111                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
 112                         break;
 113
 114                 pte = pte_offset_kernel(pmd, addr);
 115                 printk(", *pte=%0*Lx", sizeof(*pte) * 2, (u64)pte_val(*pte));
 116         } while (0);
 117
 118         printk("\n");
 119 }
 120
 121 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
 122 {
 123         unsigned index = pgd_index(address);
 124         pgd_t *pgd_k;
 125         pud_t *pud, *pud_k;
 126         pmd_t *pmd, *pmd_k;
 127
 128         pgd += index;
 129         pgd_k = init_mm.pgd + index;
 130
 131         if (!pgd_present(*pgd_k))
 132                 return NULL;
 133
 134         pud = pud_offset(pgd, address);
 135         pud_k = pud_offset(pgd_k, address);
 136         if (!pud_present(*pud_k))
 137                 return NULL;
 138
 139         if (!pud_present(*pud))
 140             set_pud(pud, *pud_k);
 141
 142         pmd = pmd_offset(pud, address);
 143         pmd_k = pmd_offset(pud_k, address);
 144         if (!pmd_present(*pmd_k))
 145                 return NULL;
 146
 147         if (!pmd_present(*pmd))
 148                 set_pmd(pmd, *pmd_k);
 149         else {
 150                 /*
 151                  * The page tables are fully synchronised so there must
 152                  * be another reason for the fault. Return NULL here to
 153                  * signal that we have not taken care of the fault.
 154                  */
 155                 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
 156                 return NULL;
 157         }
 158
 159         return pmd_k;
 160 }
 161
 162 /*
 163  * Handle a fault on the vmalloc or module mapping area
 164  */
 165 static noinline int vmalloc_fault(unsigned long address)
 166 {
 167         pgd_t *pgd_k;
 168         pmd_t *pmd_k;
 169         pte_t *pte_k;
 170
 171         /* Make sure we are in vmalloc/module/P3 area: */
 172         if (!(address >= P3SEG && address < P3_ADDR_MAX))
 173                 return -1;
 174
 175         /*
 176          * Synchronize this task's top level page-table
 177          * with the 'reference' page table.
 178          *
 179          * Do _not_ use "current" here. We might be inside
 180          * an interrupt in the middle of a task switch..
 181          */
 182         pgd_k = get_TTB();
 183         pmd_k = vmalloc_sync_one(pgd_k, address);
 184         if (!pmd_k)
 185                 return -1;
 186
 187         pte_k = pte_offset_kernel(pmd_k, address);
 188         if (!pte_present(*pte_k))
 189                 return -1;
 190
 191         return 0;
 192 }
 193
 194 static void
 195 show_fault_oops(struct pt_regs *regs, unsigned long address)
 196 {
 197         if (!oops_may_print())
 198                 return;
 199
 200         printk(KERN_ALERT "BUG: unable to handle kernel ");
 201         if (address < PAGE_SIZE)
 202                 printk(KERN_CONT "NULL pointer dereference");
 203         else
 204                 printk(KERN_CONT "paging request");
 205
 206         printk(KERN_CONT " at %08lx\n", address);
 207         printk(KERN_ALERT "PC:");
 208         printk_address(regs->pc, 1);
 209
 210         show_pte(NULL, address);
 211 }
 212
 213 static noinline void
 214 no_context(struct pt_regs *regs, unsigned long writeaccess,
 215            unsigned long address)
 216 {
 217         /* Are we prepared to handle this kernel fault?  */
 218         if (fixup_exception(regs))
 219                 return;
 220
 221         if (handle_trapped_io(regs, address))
 222                 return;
 223
 224         /*
 225          * Oops. The kernel tried to access some bad page. We'll have to
 226          * terminate things with extreme prejudice.
 227          */
 228         bust_spinlocks(1);
 229
 230         show_fault_oops(regs, address);
 231
 232         die("Oops", regs, writeaccess);
 233         bust_spinlocks(0);
 234         do_exit(SIGKILL);
 235 }
 236
 237 static void
 238 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long writeaccess,
 239                        unsigned long address, int si_code)
 240 {
 241         struct task_struct *tsk = current;
 242
 243         /* User mode accesses just cause a SIGSEGV */
 244         if (user_mode(regs)) {
 245                 /*
 246                  * It's possible to have interrupts off here:
 247                  */
 248                 local_irq_enable();
 249
 250                 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
 251
 252                 return;
 253         }
 254
 255         no_context(regs, writeaccess, address);
 256 }
 257
 258 static noinline void
 259 bad_area_nosemaphore(struct pt_regs *regs, unsigned long writeaccess,
 260                      unsigned long address)
 261 {
 262         __bad_area_nosemaphore(regs, writeaccess, address, SEGV_MAPERR);
 263 }
 264
 265 static void
 266 __bad_area(struct pt_regs *regs, unsigned long writeaccess,
 267            unsigned long address, int si_code)
 268 {
 269         struct mm_struct *mm = current->mm;
 270
 271         /*
 272          * Something tried to access memory that isn't in our memory map..
 273          * Fix it, but check if it's kernel or user first..
 274          */
 275         up_read(&mm->mmap_sem);
 276
 277         __bad_area_nosemaphore(regs, writeaccess, address, si_code);
 278 }
 279
 280 static noinline void
 281 bad_area(struct pt_regs *regs, unsigned long writeaccess, unsigned long address)
 282 {
 283         __bad_area(regs, writeaccess, address, SEGV_MAPERR);
 284 }
 285
 286 static noinline void
 287 bad_area_access_error(struct pt_regs *regs, unsigned long writeaccess,
 288                       unsigned long address)
 289 {
 290         __bad_area(regs, writeaccess, address, SEGV_ACCERR);
 291 }
 292
 293 static void out_of_memory(void)
 294 {
 295         /*
 296          * We ran out of memory, call the OOM killer, and return the userspace
 297          * (which will retry the fault, or kill us if we got oom-killed):
 298          */
 299         up_read(&current->mm->mmap_sem);
 300
 301         pagefault_out_of_memory();
 302 }
 303
 304 static void
 305 do_sigbus(struct pt_regs *regs, unsigned long writeaccess, unsigned long address)
 306 {
 307         struct task_struct *tsk = current;
 308         struct mm_struct *mm = tsk->mm;
 309
 310         up_read(&mm->mmap_sem);
 311
 312         /* Kernel mode? Handle exceptions or die: */
 313         if (!user_mode(regs))
 314                 no_context(regs, writeaccess, address);
 315
 316         force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
 317 }
 318
 319 static noinline int
 320 mm_fault_error(struct pt_regs *regs, unsigned long writeaccess,
 321                unsigned long address, unsigned int fault)
 322 {
 323         /*
 324          * Pagefault was interrupted by SIGKILL. We have no reason to
 325          * continue pagefault.
 326          */
 327         if (fatal_signal_pending(current)) {
 328                 if (!(fault & VM_FAULT_RETRY))
 329                         up_read(&current->mm->mmap_sem);
 330                 if (!user_mode(regs))
 331                         no_context(regs, writeaccess, address);
 332                 return 1;
 333         }
 334
 335         if (!(fault & VM_FAULT_ERROR))
 336                 return 0;
 337
 338         if (fault & VM_FAULT_OOM) {
 339                 /* Kernel mode? Handle exceptions or die: */
 340                 if (!user_mode(regs)) {
 341                         up_read(&current->mm->mmap_sem);
 342                         no_context(regs, writeaccess, address);
 343                         return 1;
 344                 }
 345
 346                 out_of_memory();
 347         } else {
 348                 if (fault & VM_FAULT_SIGBUS)
 349                         do_sigbus(regs, writeaccess, address);
 350                 else
 351                         BUG();
 352         }
 353
 354         return 1;
 355 }
 356
 357 static inline int access_error(int write, struct vm_area_struct *vma)
 358 {
 359         if (write) {
 360                 /* write, present and write, not present: */
 361                 if (unlikely(!(vma->vm_flags & VM_WRITE)))
 362                         return 1;
 363                 return 0;
 364         }
 365
 366         /* read, not present: */
 367         if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
 368                 return 1;
 369
 370         return 0;
 371 }
 372
 373 static int fault_in_kernel_space(unsigned long address)
 374 {
 375         return address >= TASK_SIZE;
 376 }
 377
 378 /*
 379  * This routine handles page faults.  It determines the address,
 380  * and the problem, and then passes it off to one of the appropriate
 381  * routines.
 382  */
 383 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 384                                         unsigned long writeaccess,
 385                                         unsigned long address)
 386 {
 387         unsigned long vec;
 388         struct task_struct *tsk;
 389         struct mm_struct *mm;
 390         struct vm_area_struct * vma;
 391         int fault;
 392         unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
 393                               (writeaccess ? FAULT_FLAG_WRITE : 0));
 394
 395         tsk = current;
 396         mm = tsk->mm;
 397         vec = lookup_exception_vector();
 398
 399         /*
 400          * We fault-in kernel-space virtual memory on-demand. The
 401          * 'reference' page table is init_mm.pgd.
 402          *
 403          * NOTE! We MUST NOT take any locks for this case. We may
 404          * be in an interrupt or a critical region, and should
 405          * only copy the information from the master page table,
 406          * nothing more.
 407          */
 408         if (unlikely(fault_in_kernel_space(address))) {
 409                 if (vmalloc_fault(address) >= 0)
 410                         return;
 411                 if (notify_page_fault(regs, vec))
 412                         return;
 413
 414                 bad_area_nosemaphore(regs, writeaccess, address);
 415                 return;
 416         }
 417
 418         if (unlikely(notify_page_fault(regs, vec)))
 419                 return;
 420
 421         /* Only enable interrupts if they were on before the fault */
 422         if ((regs->sr & SR_IMASK) != SR_IMASK)
 423                 local_irq_enable();
 424
 425         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 426
 427         /*
 428          * If we're in an interrupt, have no user context or are running
 429          * in an atomic region then we must not take the fault:
 430          */
 431         if (unlikely(in_atomic() || !mm)) {
 432                 bad_area_nosemaphore(regs, writeaccess, address);
 433                 return;
 434         }
 435
 436 retry:
 437         down_read(&mm->mmap_sem);
 438
 439         vma = find_vma(mm, address);
 440         if (unlikely(!vma)) {
 441                 bad_area(regs, writeaccess, address);
 442                 return;
 443         }
 444         if (likely(vma->vm_start <= address))
 445                 goto good_area;
 446         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
 447                 bad_area(regs, writeaccess, address);
 448                 return;
 449         }
 450         if (unlikely(expand_stack(vma, address))) {
 451                 bad_area(regs, writeaccess, address);
 452                 return;
 453         }
 454
 455         /*
 456          * Ok, we have a good vm_area for this memory access, so
 457          * we can handle it..
 458          */
 459 good_area:
 460         if (unlikely(access_error(writeaccess, vma))) {
 461                 bad_area_access_error(regs, writeaccess, address);
 462                 return;
 463         }
 464
 465         /*
 466          * If for any reason at all we couldn't handle the fault,
 467          * make sure we exit gracefully rather than endlessly redo
 468          * the fault.
 469          */
 470         fault = handle_mm_fault(mm, vma, address, flags);
 471
 472         if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
 473                 if (mm_fault_error(regs, writeaccess, address, fault))
 474                         return;
 475
 476         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 477                 if (fault & VM_FAULT_MAJOR) {
 478                         tsk->maj_flt++;
 479                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 480                                       regs, address);
 481                 } else {
 482                         tsk->min_flt++;
 483                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 484                                       regs, address);
 485                 }
 486                 if (fault & VM_FAULT_RETRY) {
 487                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 488
 489                         /*
 490                          * No need to up_read(&mm->mmap_sem) as we would
 491                          * have already released it in __lock_page_or_retry
 492                          * in mm/filemap.c.
 493                          */
 494                         goto retry;
 495                 }
 496         }
 497
 498         up_read(&mm->mmap_sem);
 499 }
 500
 501 /*
 502  * Called with interrupts disabled.
 503  */
 504 asmlinkage int __kprobes
 505 handle_tlbmiss(struct pt_regs *regs, unsigned long writeaccess,
 506                unsigned long address)
 507 {
 508         pgd_t *pgd;
 509         pud_t *pud;
 510         pmd_t *pmd;
 511         pte_t *pte;
 512         pte_t entry;
 513
 514         /*
 515          * We don't take page faults for P1, P2, and parts of P4, these
 516          * are always mapped, whether it be due to legacy behaviour in
 517          * 29-bit mode, or due to PMB configuration in 32-bit mode.
 518          */
 519         if (address >= P3SEG && address < P3_ADDR_MAX) {
 520                 pgd = pgd_offset_k(address);
 521         } else {
 522                 if (unlikely(address >= TASK_SIZE || !current->mm))
 523                         return 1;
 524
 525                 pgd = pgd_offset(current->mm, address);
 526         }
 527
 528         pud = pud_offset(pgd, address);
 529         if (pud_none_or_clear_bad(pud))
 530                 return 1;
 531         pmd = pmd_offset(pud, address);
 532         if (pmd_none_or_clear_bad(pmd))
 533                 return 1;
 534         pte = pte_offset_kernel(pmd, address);
 535         entry = *pte;
 536         if (unlikely(pte_none(entry) || pte_not_present(entry)))
 537                 return 1;
 538         if (unlikely(writeaccess && !pte_write(entry)))
 539                 return 1;
 540
 541         if (writeaccess)
 542                 entry = pte_mkdirty(entry);
 543         entry = pte_mkyoung(entry);
 544
 545         set_pte(pte, entry);
 546
 547 #if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SMP)
 548         /*
 549          * SH-4 does not set MMUCR.RC to the corresponding TLB entry in
 550          * the case of an initial page write exception, so we need to
 551          * flush it in order to avoid potential TLB entry duplication.
 552          */
 553         if (writeaccess == 2)
 554                 local_flush_tlb_one(get_asid(), address & PAGE_MASK);
 555 #endif
 556
 557         update_mmu_cache(NULL, address, pte);
 558
 559         return 0;
 560 }