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/sched/signal.h>
  17 #include <linux/hardirq.h>
  18 #include <linux/kprobes.h>
  19 #include <linux/perf_event.h>
  20 #include <linux/kdebug.h>
  21 #include <linux/uaccess.h>
  22 #include <asm/io_trapped.h>
  23 #include <asm/mmu_context.h>
  24 #include <asm/tlbflush.h>
  25 #include <asm/traps.h>
  26
  27 static inline int notify_page_fault(struct pt_regs *regs, int trap)
  28 {
  29         int ret = 0;
  30
  31         if (kprobes_built_in() && !user_mode(regs)) {
  32                 preempt_disable();
  33                 if (kprobe_running() && kprobe_fault_handler(regs, trap))
  34                         ret = 1;
  35                 preempt_enable();
  36         }
  37
  38         return ret;
  39 }
  40
  41 static void
  42 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
  43                      struct task_struct *tsk)
  44 {
  45         force_sig_fault(si_signo, si_code, (void __user *)address, tsk);
  46 }
  47
  48 /*
  49  * This is useful to dump out the page tables associated with
  50  * 'addr' in mm 'mm'.
  51  */
  52 static void show_pte(struct mm_struct *mm, unsigned long addr)
  53 {
  54         pgd_t *pgd;
  55
  56         if (mm) {
  57                 pgd = mm->pgd;
  58         } else {
  59                 pgd = get_TTB();
  60
  61                 if (unlikely(!pgd))
  62                         pgd = swapper_pg_dir;
  63         }
  64
  65         printk(KERN_ALERT "pgd = %p\n", pgd);
  66         pgd += pgd_index(addr);
  67         printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
  68                (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd));
  69
  70         do {
  71                 pud_t *pud;
  72                 pmd_t *pmd;
  73                 pte_t *pte;
  74
  75                 if (pgd_none(*pgd))
  76                         break;
  77
  78                 if (pgd_bad(*pgd)) {
  79                         printk("(bad)");
  80                         break;
  81                 }
  82
  83                 pud = pud_offset(pgd, addr);
  84                 if (PTRS_PER_PUD != 1)
  85                         printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
  86                                (u64)pud_val(*pud));
  87
  88                 if (pud_none(*pud))
  89                         break;
  90
  91                 if (pud_bad(*pud)) {
  92                         printk("(bad)");
  93                         break;
  94                 }
  95
  96                 pmd = pmd_offset(pud, addr);
  97                 if (PTRS_PER_PMD != 1)
  98                         printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
  99                                (u64)pmd_val(*pmd));
 100
 101                 if (pmd_none(*pmd))
 102                         break;
 103
 104                 if (pmd_bad(*pmd)) {
 105                         printk("(bad)");
 106                         break;
 107                 }
 108
 109                 /* We must not map this if we have highmem enabled */
 110                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
 111                         break;
 112
 113                 pte = pte_offset_kernel(pmd, addr);
 114                 printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
 115                        (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 #ifdef CONFIG_SH_STORE_QUEUES
 163 #define __FAULT_ADDR_LIMIT      P3_ADDR_MAX
 164 #else
 165 #define __FAULT_ADDR_LIMIT      VMALLOC_END
 166 #endif
 167
 168 /*
 169  * Handle a fault on the vmalloc or module mapping area
 170  */
 171 static noinline int vmalloc_fault(unsigned long address)
 172 {
 173         pgd_t *pgd_k;
 174         pmd_t *pmd_k;
 175         pte_t *pte_k;
 176
 177         /* Make sure we are in vmalloc/module/P3 area: */
 178         if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
 179                 return -1;
 180
 181         /*
 182          * Synchronize this task's top level page-table
 183          * with the 'reference' page table.
 184          *
 185          * Do _not_ use "current" here. We might be inside
 186          * an interrupt in the middle of a task switch..
 187          */
 188         pgd_k = get_TTB();
 189         pmd_k = vmalloc_sync_one(pgd_k, address);
 190         if (!pmd_k)
 191                 return -1;
 192
 193         pte_k = pte_offset_kernel(pmd_k, address);
 194         if (!pte_present(*pte_k))
 195                 return -1;
 196
 197         return 0;
 198 }
 199
 200 static void
 201 show_fault_oops(struct pt_regs *regs, unsigned long address)
 202 {
 203         if (!oops_may_print())
 204                 return;
 205
 206         printk(KERN_ALERT "BUG: unable to handle kernel ");
 207         if (address < PAGE_SIZE)
 208                 printk(KERN_CONT "NULL pointer dereference");
 209         else
 210                 printk(KERN_CONT "paging request");
 211
 212         printk(KERN_CONT " at %08lx\n", address);
 213         printk(KERN_ALERT "PC:");
 214         printk_address(regs->pc, 1);
 215
 216         show_pte(NULL, address);
 217 }
 218
 219 static noinline void
 220 no_context(struct pt_regs *regs, unsigned long error_code,
 221            unsigned long address)
 222 {
 223         /* Are we prepared to handle this kernel fault?  */
 224         if (fixup_exception(regs))
 225                 return;
 226
 227         if (handle_trapped_io(regs, address))
 228                 return;
 229
 230         /*
 231          * Oops. The kernel tried to access some bad page. We'll have to
 232          * terminate things with extreme prejudice.
 233          */
 234         bust_spinlocks(1);
 235
 236         show_fault_oops(regs, address);
 237
 238         die("Oops", regs, error_code);
 239         bust_spinlocks(0);
 240         do_exit(SIGKILL);
 241 }
 242
 243 static void
 244 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
 245                        unsigned long address, int si_code)
 246 {
 247         struct task_struct *tsk = current;
 248
 249         /* User mode accesses just cause a SIGSEGV */
 250         if (user_mode(regs)) {
 251                 /*
 252                  * It's possible to have interrupts off here:
 253                  */
 254                 local_irq_enable();
 255
 256                 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
 257
 258                 return;
 259         }
 260
 261         no_context(regs, error_code, address);
 262 }
 263
 264 static noinline void
 265 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
 266                      unsigned long address)
 267 {
 268         __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
 269 }
 270
 271 static void
 272 __bad_area(struct pt_regs *regs, unsigned long error_code,
 273            unsigned long address, int si_code)
 274 {
 275         struct mm_struct *mm = current->mm;
 276
 277         /*
 278          * Something tried to access memory that isn't in our memory map..
 279          * Fix it, but check if it's kernel or user first..
 280          */
 281         up_read(&mm->mmap_sem);
 282
 283         __bad_area_nosemaphore(regs, error_code, address, si_code);
 284 }
 285
 286 static noinline void
 287 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
 288 {
 289         __bad_area(regs, error_code, address, SEGV_MAPERR);
 290 }
 291
 292 static noinline void
 293 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
 294                       unsigned long address)
 295 {
 296         __bad_area(regs, error_code, address, SEGV_ACCERR);
 297 }
 298
 299 static void
 300 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
 301 {
 302         struct task_struct *tsk = current;
 303         struct mm_struct *mm = tsk->mm;
 304
 305         up_read(&mm->mmap_sem);
 306
 307         /* Kernel mode? Handle exceptions or die: */
 308         if (!user_mode(regs))
 309                 no_context(regs, error_code, address);
 310
 311         force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
 312 }
 313
 314 static noinline int
 315 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
 316                unsigned long address, vm_fault_t fault)
 317 {
 318         /*
 319          * Pagefault was interrupted by SIGKILL. We have no reason to
 320          * continue pagefault.
 321          */
 322         if (fatal_signal_pending(current)) {
 323                 if (!(fault & VM_FAULT_RETRY))
 324                         up_read(&current->mm->mmap_sem);
 325                 if (!user_mode(regs))
 326                         no_context(regs, error_code, address);
 327                 return 1;
 328         }
 329
 330         if (!(fault & VM_FAULT_ERROR))
 331                 return 0;
 332
 333         if (fault & VM_FAULT_OOM) {
 334                 /* Kernel mode? Handle exceptions or die: */
 335                 if (!user_mode(regs)) {
 336                         up_read(&current->mm->mmap_sem);
 337                         no_context(regs, error_code, address);
 338                         return 1;
 339                 }
 340                 up_read(&current->mm->mmap_sem);
 341
 342                 /*
 343                  * We ran out of memory, call the OOM killer, and return the
 344                  * userspace (which will retry the fault, or kill us if we got
 345                  * oom-killed):
 346                  */
 347                 pagefault_out_of_memory();
 348         } else {
 349                 if (fault & VM_FAULT_SIGBUS)
 350                         do_sigbus(regs, error_code, address);
 351                 else if (fault & VM_FAULT_SIGSEGV)
 352                         bad_area(regs, error_code, address);
 353                 else
 354                         BUG();
 355         }
 356
 357         return 1;
 358 }
 359
 360 static inline int access_error(int error_code, struct vm_area_struct *vma)
 361 {
 362         if (error_code & FAULT_CODE_WRITE) {
 363                 /* write, present and write, not present: */
 364                 if (unlikely(!(vma->vm_flags & VM_WRITE)))
 365                         return 1;
 366                 return 0;
 367         }
 368
 369         /* ITLB miss on NX page */
 370         if (unlikely((error_code & FAULT_CODE_ITLB) &&
 371                      !(vma->vm_flags & VM_EXEC)))
 372                 return 1;
 373
 374         /* read, not present: */
 375         if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
 376                 return 1;
 377
 378         return 0;
 379 }
 380
 381 static int fault_in_kernel_space(unsigned long address)
 382 {
 383         return address >= TASK_SIZE;
 384 }
 385
 386 /*
 387  * This routine handles page faults.  It determines the address,
 388  * and the problem, and then passes it off to one of the appropriate
 389  * routines.
 390  */
 391 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 392                                         unsigned long error_code,
 393                                         unsigned long address)
 394 {
 395         unsigned long vec;
 396         struct task_struct *tsk;
 397         struct mm_struct *mm;
 398         struct vm_area_struct * vma;
 399         vm_fault_t fault;
 400         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 401
 402         tsk = current;
 403         mm = tsk->mm;
 404         vec = lookup_exception_vector();
 405
 406         /*
 407          * We fault-in kernel-space virtual memory on-demand. The
 408          * 'reference' page table is init_mm.pgd.
 409          *
 410          * NOTE! We MUST NOT take any locks for this case. We may
 411          * be in an interrupt or a critical region, and should
 412          * only copy the information from the master page table,
 413          * nothing more.
 414          */
 415         if (unlikely(fault_in_kernel_space(address))) {
 416                 if (vmalloc_fault(address) >= 0)
 417                         return;
 418                 if (notify_page_fault(regs, vec))
 419                         return;
 420
 421                 bad_area_nosemaphore(regs, error_code, address);
 422                 return;
 423         }
 424
 425         if (unlikely(notify_page_fault(regs, vec)))
 426                 return;
 427
 428         /* Only enable interrupts if they were on before the fault */
 429         if ((regs->sr & SR_IMASK) != SR_IMASK)
 430                 local_irq_enable();
 431
 432         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 433
 434         /*
 435          * If we're in an interrupt, have no user context or are running
 436          * with pagefaults disabled then we must not take the fault:
 437          */
 438         if (unlikely(faulthandler_disabled() || !mm)) {
 439                 bad_area_nosemaphore(regs, error_code, address);
 440                 return;
 441         }
 442
 443 retry:
 444         down_read(&mm->mmap_sem);
 445
 446         vma = find_vma(mm, address);
 447         if (unlikely(!vma)) {
 448                 bad_area(regs, error_code, address);
 449                 return;
 450         }
 451         if (likely(vma->vm_start <= address))
 452                 goto good_area;
 453         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
 454                 bad_area(regs, error_code, address);
 455                 return;
 456         }
 457         if (unlikely(expand_stack(vma, address))) {
 458                 bad_area(regs, error_code, address);
 459                 return;
 460         }
 461
 462         /*
 463          * Ok, we have a good vm_area for this memory access, so
 464          * we can handle it..
 465          */
 466 good_area:
 467         if (unlikely(access_error(error_code, vma))) {
 468                 bad_area_access_error(regs, error_code, address);
 469                 return;
 470         }
 471
 472         set_thread_fault_code(error_code);
 473
 474         if (user_mode(regs))
 475                 flags |= FAULT_FLAG_USER;
 476         if (error_code & FAULT_CODE_WRITE)
 477                 flags |= FAULT_FLAG_WRITE;
 478
 479         /*
 480          * If for any reason at all we couldn't handle the fault,
 481          * make sure we exit gracefully rather than endlessly redo
 482          * the fault.
 483          */
 484         fault = handle_mm_fault(vma, address, flags);
 485
 486         if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
 487                 if (mm_fault_error(regs, error_code, address, fault))
 488                         return;
 489
 490         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 491                 if (fault & VM_FAULT_MAJOR) {
 492                         tsk->maj_flt++;
 493                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 494                                       regs, address);
 495                 } else {
 496                         tsk->min_flt++;
 497                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 498                                       regs, address);
 499                 }
 500                 if (fault & VM_FAULT_RETRY) {
 501                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 502                         flags |= FAULT_FLAG_TRIED;
 503
 504                         /*
 505                          * No need to up_read(&mm->mmap_sem) as we would
 506                          * have already released it in __lock_page_or_retry
 507                          * in mm/filemap.c.
 508                          */
 509                         goto retry;
 510                 }
 511         }
 512
 513         up_read(&mm->mmap_sem);
 514 }