1 /* 2 * linux/arch/arm/mm/fault.c 3 * 4 * Copyright (C) 1995 Linus Torvalds 5 * Modifications for ARM processor (c) 1995-2004 Russell King 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/module.h> 12 #include <linux/signal.h> 13 #include <linux/mm.h> 14 #include <linux/hardirq.h> 15 #include <linux/init.h> 16 #include <linux/kprobes.h> 17 #include <linux/uaccess.h> 18 #include <linux/page-flags.h> 19 #include <linux/sched.h> 20 #include <linux/highmem.h> 21 #include <linux/perf_event.h> 22 23 #include <asm/exception.h> 24 #include <asm/system.h> 25 #include <asm/pgtable.h> 26 #include <asm/tlbflush.h> 27 28 #include "fault.h" 29 30 /* 31 * Fault status register encodings. We steal bit 31 for our own purposes. 32 */ 33 #define FSR_LNX_PF (1 << 31) 34 #define FSR_WRITE (1 << 11) 35 #define FSR_FS4 (1 << 10) 36 #define FSR_FS3_0 (15) 37 38 static inline int fsr_fs(unsigned int fsr) 39 { 40 return (fsr & FSR_FS3_0) | (fsr & FSR_FS4) >> 6; 41 } 42 43 #ifdef CONFIG_MMU 44 45 #ifdef CONFIG_KPROBES 46 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) 47 { 48 int ret = 0; 49 50 if (!user_mode(regs)) { 51 /* kprobe_running() needs smp_processor_id() */ 52 preempt_disable(); 53 if (kprobe_running() && kprobe_fault_handler(regs, fsr)) 54 ret = 1; 55 preempt_enable(); 56 } 57 58 return ret; 59 } 60 #else 61 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) 62 { 63 return 0; 64 } 65 #endif 66 67 /* 68 * This is useful to dump out the page tables associated with 69 * 'addr' in mm 'mm'. 70 */ 71 void show_pte(struct mm_struct *mm, unsigned long addr) 72 { 73 pgd_t *pgd; 74 75 if (!mm) 76 mm = &init_mm; 77 78 printk(KERN_ALERT "pgd = %p\n", mm->pgd); 79 pgd = pgd_offset(mm, addr); 80 printk(KERN_ALERT "[%08lx] *pgd=%08llx", 81 addr, (long long)pgd_val(*pgd)); 82 83 do { 84 pud_t *pud; 85 pmd_t *pmd; 86 pte_t *pte; 87 88 if (pgd_none(*pgd)) 89 break; 90 91 if (pgd_bad(*pgd)) { 92 printk("(bad)"); 93 break; 94 } 95 96 pud = pud_offset(pgd, addr); 97 if (PTRS_PER_PUD != 1) 98 printk(", *pud=%08llx", (long long)pud_val(*pud)); 99 100 if (pud_none(*pud)) 101 break; 102 103 if (pud_bad(*pud)) { 104 printk("(bad)"); 105 break; 106 } 107 108 pmd = pmd_offset(pud, addr); 109 if (PTRS_PER_PMD != 1) 110 printk(", *pmd=%08llx", (long long)pmd_val(*pmd)); 111 112 if (pmd_none(*pmd)) 113 break; 114 115 if (pmd_bad(*pmd)) { 116 printk("(bad)"); 117 break; 118 } 119 120 /* We must not map this if we have highmem enabled */ 121 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT))) 122 break; 123 124 pte = pte_offset_map(pmd, addr); 125 printk(", *pte=%08llx", (long long)pte_val(*pte)); 126 printk(", *ppte=%08llx", 127 (long long)pte_val(pte[PTE_HWTABLE_PTRS])); 128 pte_unmap(pte); 129 } while(0); 130 131 printk("\n"); 132 } 133 #else /* CONFIG_MMU */ 134 void show_pte(struct mm_struct *mm, unsigned long addr) 135 { } 136 #endif /* CONFIG_MMU */ 137 138 /* 139 * Oops. The kernel tried to access some page that wasn't present. 140 */ 141 static void 142 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, 143 struct pt_regs *regs) 144 { 145 /* 146 * Are we prepared to handle this kernel fault? 147 */ 148 if (fixup_exception(regs)) 149 return; 150 151 /* 152 * No handler, we'll have to terminate things with extreme prejudice. 153 */ 154 bust_spinlocks(1); 155 printk(KERN_ALERT 156 "Unable to handle kernel %s at virtual address %08lx\n", 157 (addr < PAGE_SIZE) ? "NULL pointer dereference" : 158 "paging request", addr); 159 160 show_pte(mm, addr); 161 die("Oops", regs, fsr); 162 bust_spinlocks(0); 163 do_exit(SIGKILL); 164 } 165 166 /* 167 * Something tried to access memory that isn't in our memory map.. 168 * User mode accesses just cause a SIGSEGV 169 */ 170 static void 171 __do_user_fault(struct task_struct *tsk, unsigned long addr, 172 unsigned int fsr, unsigned int sig, int code, 173 struct pt_regs *regs) 174 { 175 struct siginfo si; 176 177 #ifdef CONFIG_DEBUG_USER 178 if (user_debug & UDBG_SEGV) { 179 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n", 180 tsk->comm, sig, addr, fsr); 181 show_pte(tsk->mm, addr); 182 show_regs(regs); 183 } 184 #endif 185 186 tsk->thread.address = addr; 187 tsk->thread.error_code = fsr; 188 tsk->thread.trap_no = 14; 189 si.si_signo = sig; 190 si.si_errno = 0; 191 si.si_code = code; 192 si.si_addr = (void __user *)addr; 193 force_sig_info(sig, &si, tsk); 194 } 195 196 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 197 { 198 struct task_struct *tsk = current; 199 struct mm_struct *mm = tsk->active_mm; 200 201 /* 202 * If we are in kernel mode at this point, we 203 * have no context to handle this fault with. 204 */ 205 if (user_mode(regs)) 206 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs); 207 else 208 __do_kernel_fault(mm, addr, fsr, regs); 209 } 210 211 #ifdef CONFIG_MMU 212 #define VM_FAULT_BADMAP 0x010000 213 #define VM_FAULT_BADACCESS 0x020000 214 215 /* 216 * Check that the permissions on the VMA allow for the fault which occurred. 217 * If we encountered a write fault, we must have write permission, otherwise 218 * we allow any permission. 219 */ 220 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma) 221 { 222 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC; 223 224 if (fsr & FSR_WRITE) 225 mask = VM_WRITE; 226 if (fsr & FSR_LNX_PF) 227 mask = VM_EXEC; 228 229 return vma->vm_flags & mask ? false : true; 230 } 231 232 static int __kprobes 233 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, 234 struct task_struct *tsk) 235 { 236 struct vm_area_struct *vma; 237 int fault; 238 239 vma = find_vma(mm, addr); 240 fault = VM_FAULT_BADMAP; 241 if (unlikely(!vma)) 242 goto out; 243 if (unlikely(vma->vm_start > addr)) 244 goto check_stack; 245 246 /* 247 * Ok, we have a good vm_area for this 248 * memory access, so we can handle it. 249 */ 250 good_area: 251 if (access_error(fsr, vma)) { 252 fault = VM_FAULT_BADACCESS; 253 goto out; 254 } 255 256 /* 257 * If for any reason at all we couldn't handle the fault, make 258 * sure we exit gracefully rather than endlessly redo the fault. 259 */ 260 fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & FSR_WRITE) ? FAULT_FLAG_WRITE : 0); 261 if (unlikely(fault & VM_FAULT_ERROR)) 262 return fault; 263 if (fault & VM_FAULT_MAJOR) 264 tsk->maj_flt++; 265 else 266 tsk->min_flt++; 267 return fault; 268 269 check_stack: 270 if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr)) 271 goto good_area; 272 out: 273 return fault; 274 } 275 276 static int __kprobes 277 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 278 { 279 struct task_struct *tsk; 280 struct mm_struct *mm; 281 int fault, sig, code; 282 283 if (notify_page_fault(regs, fsr)) 284 return 0; 285 286 tsk = current; 287 mm = tsk->mm; 288 289 /* Enable interrupts if they were enabled in the parent context. */ 290 if (interrupts_enabled(regs)) 291 local_irq_enable(); 292 293 /* 294 * If we're in an interrupt or have no user 295 * context, we must not take the fault.. 296 */ 297 if (in_atomic() || !mm) 298 goto no_context; 299 300 /* 301 * As per x86, we may deadlock here. However, since the kernel only 302 * validly references user space from well defined areas of the code, 303 * we can bug out early if this is from code which shouldn't. 304 */ 305 if (!down_read_trylock(&mm->mmap_sem)) { 306 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc)) 307 goto no_context; 308 down_read(&mm->mmap_sem); 309 } else { 310 /* 311 * The above down_read_trylock() might have succeeded in 312 * which case, we'll have missed the might_sleep() from 313 * down_read() 314 */ 315 might_sleep(); 316 #ifdef CONFIG_DEBUG_VM 317 if (!user_mode(regs) && 318 !search_exception_tables(regs->ARM_pc)) 319 goto no_context; 320 #endif 321 } 322 323 fault = __do_page_fault(mm, addr, fsr, tsk); 324 up_read(&mm->mmap_sem); 325 326 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr); 327 if (fault & VM_FAULT_MAJOR) 328 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, addr); 329 else if (fault & VM_FAULT_MINOR) 330 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, addr); 331 332 /* 333 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR 334 */ 335 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS)))) 336 return 0; 337 338 if (fault & VM_FAULT_OOM) { 339 /* 340 * We ran out of memory, call the OOM killer, and return to 341 * userspace (which will retry the fault, or kill us if we 342 * got oom-killed) 343 */ 344 pagefault_out_of_memory(); 345 return 0; 346 } 347 348 /* 349 * If we are in kernel mode at this point, we 350 * have no context to handle this fault with. 351 */ 352 if (!user_mode(regs)) 353 goto no_context; 354 355 if (fault & VM_FAULT_SIGBUS) { 356 /* 357 * We had some memory, but were unable to 358 * successfully fix up this page fault. 359 */ 360 sig = SIGBUS; 361 code = BUS_ADRERR; 362 } else { 363 /* 364 * Something tried to access memory that 365 * isn't in our memory map.. 366 */ 367 sig = SIGSEGV; 368 code = fault == VM_FAULT_BADACCESS ? 369 SEGV_ACCERR : SEGV_MAPERR; 370 } 371 372 __do_user_fault(tsk, addr, fsr, sig, code, regs); 373 return 0; 374 375 no_context: 376 __do_kernel_fault(mm, addr, fsr, regs); 377 return 0; 378 } 379 #else /* CONFIG_MMU */ 380 static int 381 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 382 { 383 return 0; 384 } 385 #endif /* CONFIG_MMU */ 386 387 /* 388 * First Level Translation Fault Handler 389 * 390 * We enter here because the first level page table doesn't contain 391 * a valid entry for the address. 392 * 393 * If the address is in kernel space (>= TASK_SIZE), then we are 394 * probably faulting in the vmalloc() area. 395 * 396 * If the init_task's first level page tables contains the relevant 397 * entry, we copy the it to this task. If not, we send the process 398 * a signal, fixup the exception, or oops the kernel. 399 * 400 * NOTE! We MUST NOT take any locks for this case. We may be in an 401 * interrupt or a critical region, and should only copy the information 402 * from the master page table, nothing more. 403 */ 404 #ifdef CONFIG_MMU 405 static int __kprobes 406 do_translation_fault(unsigned long addr, unsigned int fsr, 407 struct pt_regs *regs) 408 { 409 unsigned int index; 410 pgd_t *pgd, *pgd_k; 411 pud_t *pud, *pud_k; 412 pmd_t *pmd, *pmd_k; 413 414 if (addr < TASK_SIZE) 415 return do_page_fault(addr, fsr, regs); 416 417 if (user_mode(regs)) 418 goto bad_area; 419 420 index = pgd_index(addr); 421 422 /* 423 * FIXME: CP15 C1 is write only on ARMv3 architectures. 424 */ 425 pgd = cpu_get_pgd() + index; 426 pgd_k = init_mm.pgd + index; 427 428 if (pgd_none(*pgd_k)) 429 goto bad_area; 430 if (!pgd_present(*pgd)) 431 set_pgd(pgd, *pgd_k); 432 433 pud = pud_offset(pgd, addr); 434 pud_k = pud_offset(pgd_k, addr); 435 436 if (pud_none(*pud_k)) 437 goto bad_area; 438 if (!pud_present(*pud)) 439 set_pud(pud, *pud_k); 440 441 pmd = pmd_offset(pud, addr); 442 pmd_k = pmd_offset(pud_k, addr); 443 444 /* 445 * On ARM one Linux PGD entry contains two hardware entries (see page 446 * tables layout in pgtable.h). We normally guarantee that we always 447 * fill both L1 entries. But create_mapping() doesn't follow the rule. 448 * It can create inidividual L1 entries, so here we have to call 449 * pmd_none() check for the entry really corresponded to address, not 450 * for the first of pair. 451 */ 452 index = (addr >> SECTION_SHIFT) & 1; 453 if (pmd_none(pmd_k[index])) 454 goto bad_area; 455 456 copy_pmd(pmd, pmd_k); 457 return 0; 458 459 bad_area: 460 do_bad_area(addr, fsr, regs); 461 return 0; 462 } 463 #else /* CONFIG_MMU */ 464 static int 465 do_translation_fault(unsigned long addr, unsigned int fsr, 466 struct pt_regs *regs) 467 { 468 return 0; 469 } 470 #endif /* CONFIG_MMU */ 471 472 /* 473 * Some section permission faults need to be handled gracefully. 474 * They can happen due to a __{get,put}_user during an oops. 475 */ 476 static int 477 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 478 { 479 do_bad_area(addr, fsr, regs); 480 return 0; 481 } 482 483 /* 484 * This abort handler always returns "fault". 485 */ 486 static int 487 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 488 { 489 return 1; 490 } 491 492 static struct fsr_info { 493 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs); 494 int sig; 495 int code; 496 const char *name; 497 } fsr_info[] = { 498 /* 499 * The following are the standard ARMv3 and ARMv4 aborts. ARMv5 500 * defines these to be "precise" aborts. 501 */ 502 { do_bad, SIGSEGV, 0, "vector exception" }, 503 { do_bad, SIGBUS, BUS_ADRALN, "alignment exception" }, 504 { do_bad, SIGKILL, 0, "terminal exception" }, 505 { do_bad, SIGBUS, BUS_ADRALN, "alignment exception" }, 506 { do_bad, SIGBUS, 0, "external abort on linefetch" }, 507 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" }, 508 { do_bad, SIGBUS, 0, "external abort on linefetch" }, 509 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" }, 510 { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, 511 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" }, 512 { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, 513 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" }, 514 { do_bad, SIGBUS, 0, "external abort on translation" }, 515 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" }, 516 { do_bad, SIGBUS, 0, "external abort on translation" }, 517 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" }, 518 /* 519 * The following are "imprecise" aborts, which are signalled by bit 520 * 10 of the FSR, and may not be recoverable. These are only 521 * supported if the CPU abort handler supports bit 10. 522 */ 523 { do_bad, SIGBUS, 0, "unknown 16" }, 524 { do_bad, SIGBUS, 0, "unknown 17" }, 525 { do_bad, SIGBUS, 0, "unknown 18" }, 526 { do_bad, SIGBUS, 0, "unknown 19" }, 527 { do_bad, SIGBUS, 0, "lock abort" }, /* xscale */ 528 { do_bad, SIGBUS, 0, "unknown 21" }, 529 { do_bad, SIGBUS, BUS_OBJERR, "imprecise external abort" }, /* xscale */ 530 { do_bad, SIGBUS, 0, "unknown 23" }, 531 { do_bad, SIGBUS, 0, "dcache parity error" }, /* xscale */ 532 { do_bad, SIGBUS, 0, "unknown 25" }, 533 { do_bad, SIGBUS, 0, "unknown 26" }, 534 { do_bad, SIGBUS, 0, "unknown 27" }, 535 { do_bad, SIGBUS, 0, "unknown 28" }, 536 { do_bad, SIGBUS, 0, "unknown 29" }, 537 { do_bad, SIGBUS, 0, "unknown 30" }, 538 { do_bad, SIGBUS, 0, "unknown 31" } 539 }; 540 541 void __init 542 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *), 543 int sig, int code, const char *name) 544 { 545 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info)) 546 BUG(); 547 548 fsr_info[nr].fn = fn; 549 fsr_info[nr].sig = sig; 550 fsr_info[nr].code = code; 551 fsr_info[nr].name = name; 552 } 553 554 /* 555 * Dispatch a data abort to the relevant handler. 556 */ 557 asmlinkage void __exception 558 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 559 { 560 const struct fsr_info *inf = fsr_info + fsr_fs(fsr); 561 struct siginfo info; 562 563 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs)) 564 return; 565 566 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n", 567 inf->name, fsr, addr); 568 569 info.si_signo = inf->sig; 570 info.si_errno = 0; 571 info.si_code = inf->code; 572 info.si_addr = (void __user *)addr; 573 arm_notify_die("", regs, &info, fsr, 0); 574 } 575 576 577 static struct fsr_info ifsr_info[] = { 578 { do_bad, SIGBUS, 0, "unknown 0" }, 579 { do_bad, SIGBUS, 0, "unknown 1" }, 580 { do_bad, SIGBUS, 0, "debug event" }, 581 { do_bad, SIGSEGV, SEGV_ACCERR, "section access flag fault" }, 582 { do_bad, SIGBUS, 0, "unknown 4" }, 583 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" }, 584 { do_bad, SIGSEGV, SEGV_ACCERR, "page access flag fault" }, 585 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" }, 586 { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, 587 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" }, 588 { do_bad, SIGBUS, 0, "unknown 10" }, 589 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" }, 590 { do_bad, SIGBUS, 0, "external abort on translation" }, 591 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" }, 592 { do_bad, SIGBUS, 0, "external abort on translation" }, 593 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" }, 594 { do_bad, SIGBUS, 0, "unknown 16" }, 595 { do_bad, SIGBUS, 0, "unknown 17" }, 596 { do_bad, SIGBUS, 0, "unknown 18" }, 597 { do_bad, SIGBUS, 0, "unknown 19" }, 598 { do_bad, SIGBUS, 0, "unknown 20" }, 599 { do_bad, SIGBUS, 0, "unknown 21" }, 600 { do_bad, SIGBUS, 0, "unknown 22" }, 601 { do_bad, SIGBUS, 0, "unknown 23" }, 602 { do_bad, SIGBUS, 0, "unknown 24" }, 603 { do_bad, SIGBUS, 0, "unknown 25" }, 604 { do_bad, SIGBUS, 0, "unknown 26" }, 605 { do_bad, SIGBUS, 0, "unknown 27" }, 606 { do_bad, SIGBUS, 0, "unknown 28" }, 607 { do_bad, SIGBUS, 0, "unknown 29" }, 608 { do_bad, SIGBUS, 0, "unknown 30" }, 609 { do_bad, SIGBUS, 0, "unknown 31" }, 610 }; 611 612 void __init 613 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *), 614 int sig, int code, const char *name) 615 { 616 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info)) 617 BUG(); 618 619 ifsr_info[nr].fn = fn; 620 ifsr_info[nr].sig = sig; 621 ifsr_info[nr].code = code; 622 ifsr_info[nr].name = name; 623 } 624 625 asmlinkage void __exception 626 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs) 627 { 628 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr); 629 struct siginfo info; 630 631 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs)) 632 return; 633 634 printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n", 635 inf->name, ifsr, addr); 636 637 info.si_signo = inf->sig; 638 info.si_errno = 0; 639 info.si_code = inf->code; 640 info.si_addr = (void __user *)addr; 641 arm_notify_die("", regs, &info, ifsr, 0); 642 } 643 644 static int __init exceptions_init(void) 645 { 646 if (cpu_architecture() >= CPU_ARCH_ARMv6) { 647 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR, 648 "I-cache maintenance fault"); 649 } 650 651 if (cpu_architecture() >= CPU_ARCH_ARMv7) { 652 /* 653 * TODO: Access flag faults introduced in ARMv6K. 654 * Runtime check for 'K' extension is needed 655 */ 656 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR, 657 "section access flag fault"); 658 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR, 659 "section access flag fault"); 660 } 661 662 return 0; 663 } 664 665 arch_initcall(exceptions_init); 666