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