1 /* 2 * PowerPC version 3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) 4 * 5 * Derived from "arch/i386/mm/fault.c" 6 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 7 * 8 * Modified by Cort Dougan and Paul Mackerras. 9 * 10 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com) 11 * 12 * This program is free software; you can redistribute it and/or 13 * modify it under the terms of the GNU General Public License 14 * as published by the Free Software Foundation; either version 15 * 2 of the License, or (at your option) any later version. 16 */ 17 18 #include <linux/signal.h> 19 #include <linux/sched.h> 20 #include <linux/sched/task_stack.h> 21 #include <linux/kernel.h> 22 #include <linux/errno.h> 23 #include <linux/string.h> 24 #include <linux/types.h> 25 #include <linux/pagemap.h> 26 #include <linux/ptrace.h> 27 #include <linux/mman.h> 28 #include <linux/mm.h> 29 #include <linux/interrupt.h> 30 #include <linux/highmem.h> 31 #include <linux/extable.h> 32 #include <linux/kprobes.h> 33 #include <linux/kdebug.h> 34 #include <linux/perf_event.h> 35 #include <linux/ratelimit.h> 36 #include <linux/context_tracking.h> 37 #include <linux/hugetlb.h> 38 #include <linux/uaccess.h> 39 40 #include <asm/firmware.h> 41 #include <asm/page.h> 42 #include <asm/pgtable.h> 43 #include <asm/mmu.h> 44 #include <asm/mmu_context.h> 45 #include <asm/siginfo.h> 46 #include <asm/debug.h> 47 48 static inline bool notify_page_fault(struct pt_regs *regs) 49 { 50 bool ret = false; 51 52 #ifdef CONFIG_KPROBES 53 /* kprobe_running() needs smp_processor_id() */ 54 if (!user_mode(regs)) { 55 preempt_disable(); 56 if (kprobe_running() && kprobe_fault_handler(regs, 11)) 57 ret = true; 58 preempt_enable(); 59 } 60 #endif /* CONFIG_KPROBES */ 61 62 if (unlikely(debugger_fault_handler(regs))) 63 ret = true; 64 65 return ret; 66 } 67 68 /* 69 * Check whether the instruction inst is a store using 70 * an update addressing form which will update r1. 71 */ 72 static bool store_updates_sp(unsigned int inst) 73 { 74 /* check for 1 in the rA field */ 75 if (((inst >> 16) & 0x1f) != 1) 76 return false; 77 /* check major opcode */ 78 switch (inst >> 26) { 79 case OP_STWU: 80 case OP_STBU: 81 case OP_STHU: 82 case OP_STFSU: 83 case OP_STFDU: 84 return true; 85 case OP_STD: /* std or stdu */ 86 return (inst & 3) == 1; 87 case OP_31: 88 /* check minor opcode */ 89 switch ((inst >> 1) & 0x3ff) { 90 case OP_31_XOP_STDUX: 91 case OP_31_XOP_STWUX: 92 case OP_31_XOP_STBUX: 93 case OP_31_XOP_STHUX: 94 case OP_31_XOP_STFSUX: 95 case OP_31_XOP_STFDUX: 96 return true; 97 } 98 } 99 return false; 100 } 101 /* 102 * do_page_fault error handling helpers 103 */ 104 105 static int 106 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code, 107 int pkey) 108 { 109 /* 110 * If we are in kernel mode, bail out with a SEGV, this will 111 * be caught by the assembly which will restore the non-volatile 112 * registers before calling bad_page_fault() 113 */ 114 if (!user_mode(regs)) 115 return SIGSEGV; 116 117 _exception_pkey(SIGSEGV, regs, si_code, address, pkey); 118 119 return 0; 120 } 121 122 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address) 123 { 124 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR, 0); 125 } 126 127 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code, 128 int pkey) 129 { 130 struct mm_struct *mm = current->mm; 131 132 /* 133 * Something tried to access memory that isn't in our memory map.. 134 * Fix it, but check if it's kernel or user first.. 135 */ 136 up_read(&mm->mmap_sem); 137 138 return __bad_area_nosemaphore(regs, address, si_code, pkey); 139 } 140 141 static noinline int bad_area(struct pt_regs *regs, unsigned long address) 142 { 143 return __bad_area(regs, address, SEGV_MAPERR, 0); 144 } 145 146 static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address, 147 int pkey) 148 { 149 return __bad_area_nosemaphore(regs, address, SEGV_PKUERR, pkey); 150 } 151 152 static noinline int bad_access(struct pt_regs *regs, unsigned long address) 153 { 154 return __bad_area(regs, address, SEGV_ACCERR, 0); 155 } 156 157 static int do_sigbus(struct pt_regs *regs, unsigned long address, 158 unsigned int fault) 159 { 160 siginfo_t info; 161 unsigned int lsb = 0; 162 163 if (!user_mode(regs)) 164 return SIGBUS; 165 166 current->thread.trap_nr = BUS_ADRERR; 167 clear_siginfo(&info); 168 info.si_signo = SIGBUS; 169 info.si_errno = 0; 170 info.si_code = BUS_ADRERR; 171 info.si_addr = (void __user *)address; 172 #ifdef CONFIG_MEMORY_FAILURE 173 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 174 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 175 current->comm, current->pid, address); 176 info.si_code = BUS_MCEERR_AR; 177 } 178 179 if (fault & VM_FAULT_HWPOISON_LARGE) 180 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 181 if (fault & VM_FAULT_HWPOISON) 182 lsb = PAGE_SHIFT; 183 #endif 184 info.si_addr_lsb = lsb; 185 force_sig_info(SIGBUS, &info, current); 186 return 0; 187 } 188 189 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault) 190 { 191 /* 192 * Kernel page fault interrupted by SIGKILL. We have no reason to 193 * continue processing. 194 */ 195 if (fatal_signal_pending(current) && !user_mode(regs)) 196 return SIGKILL; 197 198 /* Out of memory */ 199 if (fault & VM_FAULT_OOM) { 200 /* 201 * We ran out of memory, or some other thing happened to us that 202 * made us unable to handle the page fault gracefully. 203 */ 204 if (!user_mode(regs)) 205 return SIGSEGV; 206 pagefault_out_of_memory(); 207 } else { 208 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 209 VM_FAULT_HWPOISON_LARGE)) 210 return do_sigbus(regs, addr, fault); 211 else if (fault & VM_FAULT_SIGSEGV) 212 return bad_area_nosemaphore(regs, addr); 213 else 214 BUG(); 215 } 216 return 0; 217 } 218 219 /* Is this a bad kernel fault ? */ 220 static bool bad_kernel_fault(bool is_exec, unsigned long error_code, 221 unsigned long address) 222 { 223 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT))) { 224 printk_ratelimited(KERN_CRIT "kernel tried to execute" 225 " exec-protected page (%lx) -" 226 "exploit attempt? (uid: %d)\n", 227 address, from_kuid(&init_user_ns, 228 current_uid())); 229 } 230 return is_exec || (address >= TASK_SIZE); 231 } 232 233 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address, 234 struct vm_area_struct *vma, unsigned int flags, 235 bool *must_retry) 236 { 237 /* 238 * N.B. The POWER/Open ABI allows programs to access up to 239 * 288 bytes below the stack pointer. 240 * The kernel signal delivery code writes up to about 1.5kB 241 * below the stack pointer (r1) before decrementing it. 242 * The exec code can write slightly over 640kB to the stack 243 * before setting the user r1. Thus we allow the stack to 244 * expand to 1MB without further checks. 245 */ 246 if (address + 0x100000 < vma->vm_end) { 247 unsigned int __user *nip = (unsigned int __user *)regs->nip; 248 /* get user regs even if this fault is in kernel mode */ 249 struct pt_regs *uregs = current->thread.regs; 250 if (uregs == NULL) 251 return true; 252 253 /* 254 * A user-mode access to an address a long way below 255 * the stack pointer is only valid if the instruction 256 * is one which would update the stack pointer to the 257 * address accessed if the instruction completed, 258 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb 259 * (or the byte, halfword, float or double forms). 260 * 261 * If we don't check this then any write to the area 262 * between the last mapped region and the stack will 263 * expand the stack rather than segfaulting. 264 */ 265 if (address + 2048 >= uregs->gpr[1]) 266 return false; 267 268 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) && 269 access_ok(VERIFY_READ, nip, sizeof(*nip))) { 270 unsigned int inst; 271 int res; 272 273 pagefault_disable(); 274 res = __get_user_inatomic(inst, nip); 275 pagefault_enable(); 276 if (!res) 277 return !store_updates_sp(inst); 278 *must_retry = true; 279 } 280 return true; 281 } 282 return false; 283 } 284 285 static bool access_error(bool is_write, bool is_exec, 286 struct vm_area_struct *vma) 287 { 288 /* 289 * Allow execution from readable areas if the MMU does not 290 * provide separate controls over reading and executing. 291 * 292 * Note: That code used to not be enabled for 4xx/BookE. 293 * It is now as I/D cache coherency for these is done at 294 * set_pte_at() time and I see no reason why the test 295 * below wouldn't be valid on those processors. This -may- 296 * break programs compiled with a really old ABI though. 297 */ 298 if (is_exec) { 299 return !(vma->vm_flags & VM_EXEC) && 300 (cpu_has_feature(CPU_FTR_NOEXECUTE) || 301 !(vma->vm_flags & (VM_READ | VM_WRITE))); 302 } 303 304 if (is_write) { 305 if (unlikely(!(vma->vm_flags & VM_WRITE))) 306 return true; 307 return false; 308 } 309 310 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 311 return true; 312 /* 313 * We should ideally do the vma pkey access check here. But in the 314 * fault path, handle_mm_fault() also does the same check. To avoid 315 * these multiple checks, we skip it here and handle access error due 316 * to pkeys later. 317 */ 318 return false; 319 } 320 321 #ifdef CONFIG_PPC_SMLPAR 322 static inline void cmo_account_page_fault(void) 323 { 324 if (firmware_has_feature(FW_FEATURE_CMO)) { 325 u32 page_ins; 326 327 preempt_disable(); 328 page_ins = be32_to_cpu(get_lppaca()->page_ins); 329 page_ins += 1 << PAGE_FACTOR; 330 get_lppaca()->page_ins = cpu_to_be32(page_ins); 331 preempt_enable(); 332 } 333 } 334 #else 335 static inline void cmo_account_page_fault(void) { } 336 #endif /* CONFIG_PPC_SMLPAR */ 337 338 #ifdef CONFIG_PPC_STD_MMU 339 static void sanity_check_fault(bool is_write, unsigned long error_code) 340 { 341 /* 342 * For hash translation mode, we should never get a 343 * PROTFAULT. Any update to pte to reduce access will result in us 344 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE 345 * fault instead of DSISR_PROTFAULT. 346 * 347 * A pte update to relax the access will not result in a hash page table 348 * entry invalidate and hence can result in DSISR_PROTFAULT. 349 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have 350 * the special !is_write in the below conditional. 351 * 352 * For platforms that doesn't supports coherent icache and do support 353 * per page noexec bit, we do setup things such that we do the 354 * sync between D/I cache via fault. But that is handled via low level 355 * hash fault code (hash_page_do_lazy_icache()) and we should not reach 356 * here in such case. 357 * 358 * For wrong access that can result in PROTFAULT, the above vma->vm_flags 359 * check should handle those and hence we should fall to the bad_area 360 * handling correctly. 361 * 362 * For embedded with per page exec support that doesn't support coherent 363 * icache we do get PROTFAULT and we handle that D/I cache sync in 364 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON 365 * is conditional for server MMU. 366 * 367 * For radix, we can get prot fault for autonuma case, because radix 368 * page table will have them marked noaccess for user. 369 */ 370 if (!radix_enabled() && !is_write) 371 WARN_ON_ONCE(error_code & DSISR_PROTFAULT); 372 } 373 #else 374 static void sanity_check_fault(bool is_write, unsigned long error_code) { } 375 #endif /* CONFIG_PPC_STD_MMU */ 376 377 /* 378 * Define the correct "is_write" bit in error_code based 379 * on the processor family 380 */ 381 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) 382 #define page_fault_is_write(__err) ((__err) & ESR_DST) 383 #define page_fault_is_bad(__err) (0) 384 #else 385 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE) 386 #if defined(CONFIG_PPC_8xx) 387 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G) 388 #elif defined(CONFIG_PPC64) 389 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S) 390 #else 391 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S) 392 #endif 393 #endif 394 395 /* 396 * For 600- and 800-family processors, the error_code parameter is DSISR 397 * for a data fault, SRR1 for an instruction fault. For 400-family processors 398 * the error_code parameter is ESR for a data fault, 0 for an instruction 399 * fault. 400 * For 64-bit processors, the error_code parameter is 401 * - DSISR for a non-SLB data access fault, 402 * - SRR1 & 0x08000000 for a non-SLB instruction access fault 403 * - 0 any SLB fault. 404 * 405 * The return value is 0 if the fault was handled, or the signal 406 * number if this is a kernel fault that can't be handled here. 407 */ 408 static int __do_page_fault(struct pt_regs *regs, unsigned long address, 409 unsigned long error_code) 410 { 411 struct vm_area_struct * vma; 412 struct mm_struct *mm = current->mm; 413 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 414 int is_exec = TRAP(regs) == 0x400; 415 int is_user = user_mode(regs); 416 int is_write = page_fault_is_write(error_code); 417 int fault, major = 0; 418 bool must_retry = false; 419 420 if (notify_page_fault(regs)) 421 return 0; 422 423 if (unlikely(page_fault_is_bad(error_code))) { 424 if (is_user) { 425 _exception(SIGBUS, regs, BUS_OBJERR, address); 426 return 0; 427 } 428 return SIGBUS; 429 } 430 431 /* Additional sanity check(s) */ 432 sanity_check_fault(is_write, error_code); 433 434 /* 435 * The kernel should never take an execute fault nor should it 436 * take a page fault to a kernel address. 437 */ 438 if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address))) 439 return SIGSEGV; 440 441 /* 442 * If we're in an interrupt, have no user context or are running 443 * in a region with pagefaults disabled then we must not take the fault 444 */ 445 if (unlikely(faulthandler_disabled() || !mm)) { 446 if (is_user) 447 printk_ratelimited(KERN_ERR "Page fault in user mode" 448 " with faulthandler_disabled()=%d" 449 " mm=%p\n", 450 faulthandler_disabled(), mm); 451 return bad_area_nosemaphore(regs, address); 452 } 453 454 /* We restore the interrupt state now */ 455 if (!arch_irq_disabled_regs(regs)) 456 local_irq_enable(); 457 458 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 459 460 if (error_code & DSISR_KEYFAULT) 461 return bad_key_fault_exception(regs, address, 462 get_mm_addr_key(mm, address)); 463 464 /* 465 * We want to do this outside mmap_sem, because reading code around nip 466 * can result in fault, which will cause a deadlock when called with 467 * mmap_sem held 468 */ 469 if (is_user) 470 flags |= FAULT_FLAG_USER; 471 if (is_write) 472 flags |= FAULT_FLAG_WRITE; 473 if (is_exec) 474 flags |= FAULT_FLAG_INSTRUCTION; 475 476 /* When running in the kernel we expect faults to occur only to 477 * addresses in user space. All other faults represent errors in the 478 * kernel and should generate an OOPS. Unfortunately, in the case of an 479 * erroneous fault occurring in a code path which already holds mmap_sem 480 * we will deadlock attempting to validate the fault against the 481 * address space. Luckily the kernel only validly references user 482 * space from well defined areas of code, which are listed in the 483 * exceptions table. 484 * 485 * As the vast majority of faults will be valid we will only perform 486 * the source reference check when there is a possibility of a deadlock. 487 * Attempt to lock the address space, if we cannot we then validate the 488 * source. If this is invalid we can skip the address space check, 489 * thus avoiding the deadlock. 490 */ 491 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 492 if (!is_user && !search_exception_tables(regs->nip)) 493 return bad_area_nosemaphore(regs, address); 494 495 retry: 496 down_read(&mm->mmap_sem); 497 } else { 498 /* 499 * The above down_read_trylock() might have succeeded in 500 * which case we'll have missed the might_sleep() from 501 * down_read(): 502 */ 503 might_sleep(); 504 } 505 506 vma = find_vma(mm, address); 507 if (unlikely(!vma)) 508 return bad_area(regs, address); 509 if (likely(vma->vm_start <= address)) 510 goto good_area; 511 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) 512 return bad_area(regs, address); 513 514 /* The stack is being expanded, check if it's valid */ 515 if (unlikely(bad_stack_expansion(regs, address, vma, flags, 516 &must_retry))) { 517 if (!must_retry) 518 return bad_area(regs, address); 519 520 up_read(&mm->mmap_sem); 521 if (fault_in_pages_readable((const char __user *)regs->nip, 522 sizeof(unsigned int))) 523 return bad_area_nosemaphore(regs, address); 524 goto retry; 525 } 526 527 /* Try to expand it */ 528 if (unlikely(expand_stack(vma, address))) 529 return bad_area(regs, address); 530 531 good_area: 532 if (unlikely(access_error(is_write, is_exec, vma))) 533 return bad_access(regs, address); 534 535 /* 536 * If for any reason at all we couldn't handle the fault, 537 * make sure we exit gracefully rather than endlessly redo 538 * the fault. 539 */ 540 fault = handle_mm_fault(vma, address, flags); 541 542 #ifdef CONFIG_PPC_MEM_KEYS 543 /* 544 * we skipped checking for access error due to key earlier. 545 * Check that using handle_mm_fault error return. 546 */ 547 if (unlikely(fault & VM_FAULT_SIGSEGV) && 548 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) { 549 550 int pkey = vma_pkey(vma); 551 552 up_read(&mm->mmap_sem); 553 return bad_key_fault_exception(regs, address, pkey); 554 } 555 #endif /* CONFIG_PPC_MEM_KEYS */ 556 557 major |= fault & VM_FAULT_MAJOR; 558 559 /* 560 * Handle the retry right now, the mmap_sem has been released in that 561 * case. 562 */ 563 if (unlikely(fault & VM_FAULT_RETRY)) { 564 /* We retry only once */ 565 if (flags & FAULT_FLAG_ALLOW_RETRY) { 566 /* 567 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk 568 * of starvation. 569 */ 570 flags &= ~FAULT_FLAG_ALLOW_RETRY; 571 flags |= FAULT_FLAG_TRIED; 572 if (!fatal_signal_pending(current)) 573 goto retry; 574 } 575 576 /* 577 * User mode? Just return to handle the fatal exception otherwise 578 * return to bad_page_fault 579 */ 580 return is_user ? 0 : SIGBUS; 581 } 582 583 up_read(¤t->mm->mmap_sem); 584 585 if (unlikely(fault & VM_FAULT_ERROR)) 586 return mm_fault_error(regs, address, fault); 587 588 /* 589 * Major/minor page fault accounting. 590 */ 591 if (major) { 592 current->maj_flt++; 593 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 594 cmo_account_page_fault(); 595 } else { 596 current->min_flt++; 597 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 598 } 599 return 0; 600 } 601 NOKPROBE_SYMBOL(__do_page_fault); 602 603 int do_page_fault(struct pt_regs *regs, unsigned long address, 604 unsigned long error_code) 605 { 606 enum ctx_state prev_state = exception_enter(); 607 int rc = __do_page_fault(regs, address, error_code); 608 exception_exit(prev_state); 609 return rc; 610 } 611 NOKPROBE_SYMBOL(do_page_fault); 612 613 /* 614 * bad_page_fault is called when we have a bad access from the kernel. 615 * It is called from the DSI and ISI handlers in head.S and from some 616 * of the procedures in traps.c. 617 */ 618 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) 619 { 620 const struct exception_table_entry *entry; 621 622 /* Are we prepared to handle this fault? */ 623 if ((entry = search_exception_tables(regs->nip)) != NULL) { 624 regs->nip = extable_fixup(entry); 625 return; 626 } 627 628 /* kernel has accessed a bad area */ 629 630 switch (TRAP(regs)) { 631 case 0x300: 632 case 0x380: 633 printk(KERN_ALERT "Unable to handle kernel paging request for " 634 "data at address 0x%08lx\n", regs->dar); 635 break; 636 case 0x400: 637 case 0x480: 638 printk(KERN_ALERT "Unable to handle kernel paging request for " 639 "instruction fetch\n"); 640 break; 641 case 0x600: 642 printk(KERN_ALERT "Unable to handle kernel paging request for " 643 "unaligned access at address 0x%08lx\n", regs->dar); 644 break; 645 default: 646 printk(KERN_ALERT "Unable to handle kernel paging request for " 647 "unknown fault\n"); 648 break; 649 } 650 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", 651 regs->nip); 652 653 if (task_stack_end_corrupted(current)) 654 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); 655 656 die("Kernel access of bad area", regs, sig); 657 } 658