arch/sh/mm/fault.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                 if (unlikely(!pgd))
  67                         pgd = swapper_pg_dir;
  68         }
  69
  70         printk(KERN_ALERT "pgd = %p\n", pgd);
  71         pgd += pgd_index(addr);
  72         printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
  73                (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd));
  74
  75         do {
  76                 pud_t *pud;
  77                 pmd_t *pmd;
  78                 pte_t *pte;
  79
  80                 if (pgd_none(*pgd))
  81                         break;
  82
  83                 if (pgd_bad(*pgd)) {
  84                         printk("(bad)");
  85                         break;
  86                 }
  87
  88                 pud = pud_offset(pgd, addr);
  89                 if (PTRS_PER_PUD != 1)
  90                         printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
  91                                (u64)pud_val(*pud));
  92
  93                 if (pud_none(*pud))
  94                         break;
  95
  96                 if (pud_bad(*pud)) {
  97                         printk("(bad)");
  98                         break;
  99                 }
 100
 101                 pmd = pmd_offset(pud, addr);
 102                 if (PTRS_PER_PMD != 1)
 103                         printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
 104                                (u64)pmd_val(*pmd));
 105
 106                 if (pmd_none(*pmd))
 107                         break;
 108
 109                 if (pmd_bad(*pmd)) {
 110                         printk("(bad)");
 111                         break;
 112                 }
 113
 114                 /* We must not map this if we have highmem enabled */
 115                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
 116                         break;
 117
 118                 pte = pte_offset_kernel(pmd, addr);
 119                 printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
 120                        (u64)pte_val(*pte));
 121         } while (0);
 122
 123         printk("\n");
 124 }
 125
 126 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
 127 {
 128         unsigned index = pgd_index(address);
 129         pgd_t *pgd_k;
 130         pud_t *pud, *pud_k;
 131         pmd_t *pmd, *pmd_k;
 132
 133         pgd += index;
 134         pgd_k = init_mm.pgd + index;
 135
 136         if (!pgd_present(*pgd_k))
 137                 return NULL;
 138
 139         pud = pud_offset(pgd, address);
 140         pud_k = pud_offset(pgd_k, address);
 141         if (!pud_present(*pud_k))
 142                 return NULL;
 143
 144         if (!pud_present(*pud))
 145             set_pud(pud, *pud_k);
 146
 147         pmd = pmd_offset(pud, address);
 148         pmd_k = pmd_offset(pud_k, address);
 149         if (!pmd_present(*pmd_k))
 150                 return NULL;
 151
 152         if (!pmd_present(*pmd))
 153                 set_pmd(pmd, *pmd_k);
 154         else {
 155                 /*
 156                  * The page tables are fully synchronised so there must
 157                  * be another reason for the fault. Return NULL here to
 158                  * signal that we have not taken care of the fault.
 159                  */
 160                 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
 161                 return NULL;
 162         }
 163
 164         return pmd_k;
 165 }
 166
 167 #ifdef CONFIG_SH_STORE_QUEUES
 168 #define __FAULT_ADDR_LIMIT      P3_ADDR_MAX
 169 #else
 170 #define __FAULT_ADDR_LIMIT      VMALLOC_END
 171 #endif
 172
 173 /*
 174  * Handle a fault on the vmalloc or module mapping area
 175  */
 176 static noinline int vmalloc_fault(unsigned long address)
 177 {
 178         pgd_t *pgd_k;
 179         pmd_t *pmd_k;
 180         pte_t *pte_k;
 181
 182         /* Make sure we are in vmalloc/module/P3 area: */
 183         if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
 184                 return -1;
 185
 186         /*
 187          * Synchronize this task's top level page-table
 188          * with the 'reference' page table.
 189          *
 190          * Do _not_ use "current" here. We might be inside
 191          * an interrupt in the middle of a task switch..
 192          */
 193         pgd_k = get_TTB();
 194         pmd_k = vmalloc_sync_one(pgd_k, address);
 195         if (!pmd_k)
 196                 return -1;
 197
 198         pte_k = pte_offset_kernel(pmd_k, address);
 199         if (!pte_present(*pte_k))
 200                 return -1;
 201
 202         return 0;
 203 }
 204
 205 static void
 206 show_fault_oops(struct pt_regs *regs, unsigned long address)
 207 {
 208         if (!oops_may_print())
 209                 return;
 210
 211         printk(KERN_ALERT "BUG: unable to handle kernel ");
 212         if (address < PAGE_SIZE)
 213                 printk(KERN_CONT "NULL pointer dereference");
 214         else
 215                 printk(KERN_CONT "paging request");
 216
 217         printk(KERN_CONT " at %08lx\n", address);
 218         printk(KERN_ALERT "PC:");
 219         printk_address(regs->pc, 1);
 220
 221         show_pte(NULL, address);
 222 }
 223
 224 static noinline void
 225 no_context(struct pt_regs *regs, unsigned long error_code,
 226            unsigned long address)
 227 {
 228         /* Are we prepared to handle this kernel fault?  */
 229         if (fixup_exception(regs))
 230                 return;
 231
 232         if (handle_trapped_io(regs, address))
 233                 return;
 234
 235         /*
 236          * Oops. The kernel tried to access some bad page. We'll have to
 237          * terminate things with extreme prejudice.
 238          */
 239         bust_spinlocks(1);
 240
 241         show_fault_oops(regs, address);
 242
 243         die("Oops", regs, error_code);
 244         bust_spinlocks(0);
 245         do_exit(SIGKILL);
 246 }
 247
 248 static void
 249 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
 250                        unsigned long address, int si_code)
 251 {
 252         struct task_struct *tsk = current;
 253
 254         /* User mode accesses just cause a SIGSEGV */
 255         if (user_mode(regs)) {
 256                 /*
 257                  * It's possible to have interrupts off here:
 258                  */
 259                 local_irq_enable();
 260
 261                 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
 262
 263                 return;
 264         }
 265
 266         no_context(regs, error_code, address);
 267 }
 268
 269 static noinline void
 270 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
 271                      unsigned long address)
 272 {
 273         __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
 274 }
 275
 276 static void
 277 __bad_area(struct pt_regs *regs, unsigned long error_code,
 278            unsigned long address, int si_code)
 279 {
 280         struct mm_struct *mm = current->mm;
 281
 282         /*
 283          * Something tried to access memory that isn't in our memory map..
 284          * Fix it, but check if it's kernel or user first..
 285          */
 286         up_read(&mm->mmap_sem);
 287
 288         __bad_area_nosemaphore(regs, error_code, address, si_code);
 289 }
 290
 291 static noinline void
 292 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
 293 {
 294         __bad_area(regs, error_code, address, SEGV_MAPERR);
 295 }
 296
 297 static noinline void
 298 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
 299                       unsigned long address)
 300 {
 301         __bad_area(regs, error_code, address, SEGV_ACCERR);
 302 }
 303
 304 static void out_of_memory(void)
 305 {
 306         /*
 307          * We ran out of memory, call the OOM killer, and return the userspace
 308          * (which will retry the fault, or kill us if we got oom-killed):
 309          */
 310         up_read(&current->mm->mmap_sem);
 311
 312         pagefault_out_of_memory();
 313 }
 314
 315 static void
 316 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
 317 {
 318         struct task_struct *tsk = current;
 319         struct mm_struct *mm = tsk->mm;
 320
 321         up_read(&mm->mmap_sem);
 322
 323         /* Kernel mode? Handle exceptions or die: */
 324         if (!user_mode(regs))
 325                 no_context(regs, error_code, address);
 326
 327         force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
 328 }
 329
 330 static noinline int
 331 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
 332                unsigned long address, unsigned int fault)
 333 {
 334         /*
 335          * Pagefault was interrupted by SIGKILL. We have no reason to
 336          * continue pagefault.
 337          */
 338         if (fatal_signal_pending(current)) {
 339                 if (!(fault & VM_FAULT_RETRY))
 340                         up_read(&current->mm->mmap_sem);
 341                 if (!user_mode(regs))
 342                         no_context(regs, error_code, address);
 343                 return 1;
 344         }
 345
 346         if (!(fault & VM_FAULT_ERROR))
 347                 return 0;
 348
 349         if (fault & VM_FAULT_OOM) {
 350                 /* Kernel mode? Handle exceptions or die: */
 351                 if (!user_mode(regs)) {
 352                         up_read(&current->mm->mmap_sem);
 353                         no_context(regs, error_code, address);
 354                         return 1;
 355                 }
 356
 357                 out_of_memory();
 358         } else {
 359                 if (fault & VM_FAULT_SIGBUS)
 360                         do_sigbus(regs, error_code, address);
 361                 else
 362                         BUG();
 363         }
 364
 365         return 1;
 366 }
 367
 368 static inline int access_error(int error_code, struct vm_area_struct *vma)
 369 {
 370         if (error_code & FAULT_CODE_WRITE) {
 371                 /* write, present and write, not present: */
 372                 if (unlikely(!(vma->vm_flags & VM_WRITE)))
 373                         return 1;
 374                 return 0;
 375         }
 376
 377         /* ITLB miss on NX page */
 378         if (unlikely((error_code & FAULT_CODE_ITLB) &&
 379                      !(vma->vm_flags & VM_EXEC)))
 380                 return 1;
 381
 382         /* read, not present: */
 383         if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
 384                 return 1;
 385
 386         return 0;
 387 }
 388
 389 static int fault_in_kernel_space(unsigned long address)
 390 {
 391         return address >= TASK_SIZE;
 392 }
 393
 394 /*
 395  * This routine handles page faults.  It determines the address,
 396  * and the problem, and then passes it off to one of the appropriate
 397  * routines.
 398  */
 399 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 400                                         unsigned long error_code,
 401                                         unsigned long address)
 402 {
 403         unsigned long vec;
 404         struct task_struct *tsk;
 405         struct mm_struct *mm;
 406         struct vm_area_struct * vma;
 407         int fault;
 408         int write = error_code & FAULT_CODE_WRITE;
 409         unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
 410                               (write ? FAULT_FLAG_WRITE : 0));
 411
 412         tsk = current;
 413         mm = tsk->mm;
 414         vec = lookup_exception_vector();
 415
 416         /*
 417          * We fault-in kernel-space virtual memory on-demand. The
 418          * 'reference' page table is init_mm.pgd.
 419          *
 420          * NOTE! We MUST NOT take any locks for this case. We may
 421          * be in an interrupt or a critical region, and should
 422          * only copy the information from the master page table,
 423          * nothing more.
 424          */
 425         if (unlikely(fault_in_kernel_space(address))) {
 426                 if (vmalloc_fault(address) >= 0)
 427                         return;
 428                 if (notify_page_fault(regs, vec))
 429                         return;
 430
 431                 bad_area_nosemaphore(regs, error_code, address);
 432                 return;
 433         }
 434
 435         if (unlikely(notify_page_fault(regs, vec)))
 436                 return;
 437
 438         /* Only enable interrupts if they were on before the fault */
 439         if ((regs->sr & SR_IMASK) != SR_IMASK)
 440                 local_irq_enable();
 441
 442         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 443
 444         /*
 445          * If we're in an interrupt, have no user context or are running
 446          * in an atomic region then we must not take the fault:
 447          */
 448         if (unlikely(in_atomic() || !mm)) {
 449                 bad_area_nosemaphore(regs, error_code, address);
 450                 return;
 451         }
 452
 453 retry:
 454         down_read(&mm->mmap_sem);
 455
 456         vma = find_vma(mm, address);
 457         if (unlikely(!vma)) {
 458                 bad_area(regs, error_code, address);
 459                 return;
 460         }
 461         if (likely(vma->vm_start <= address))
 462                 goto good_area;
 463         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
 464                 bad_area(regs, error_code, address);
 465                 return;
 466         }
 467         if (unlikely(expand_stack(vma, address))) {
 468                 bad_area(regs, error_code, address);
 469                 return;
 470         }
 471
 472         /*
 473          * Ok, we have a good vm_area for this memory access, so
 474          * we can handle it..
 475          */
 476 good_area:
 477         if (unlikely(access_error(error_code, vma))) {
 478                 bad_area_access_error(regs, error_code, address);
 479                 return;
 480         }
 481
 482         set_thread_fault_code(error_code);
 483
 484         /*
 485          * If for any reason at all we couldn't handle the fault,
 486          * make sure we exit gracefully rather than endlessly redo
 487          * the fault.
 488          */
 489         fault = handle_mm_fault(mm, vma, address, flags);
 490
 491         if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
 492                 if (mm_fault_error(regs, error_code, address, fault))
 493                         return;
 494
 495         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 496                 if (fault & VM_FAULT_MAJOR) {
 497                         tsk->maj_flt++;
 498                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 499                                       regs, address);
 500                 } else {
 501                         tsk->min_flt++;
 502                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 503                                       regs, address);
 504                 }
 505                 if (fault & VM_FAULT_RETRY) {
 506                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 507
 508                         /*
 509                          * No need to up_read(&mm->mmap_sem) as we would
 510                          * have already released it in __lock_page_or_retry
 511                          * in mm/filemap.c.
 512                          */
 513                         goto retry;
 514                 }
 515         }
 516
 517         up_read(&mm->mmap_sem);
 518 }