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/ptrace.h> 26 #include <linux/mman.h> 27 #include <linux/mm.h> 28 #include <linux/interrupt.h> 29 #include <linux/highmem.h> 30 #include <linux/extable.h> 31 #include <linux/kprobes.h> 32 #include <linux/kdebug.h> 33 #include <linux/perf_event.h> 34 #include <linux/ratelimit.h> 35 #include <linux/context_tracking.h> 36 #include <linux/hugetlb.h> 37 #include <linux/uaccess.h> 38 39 #include <asm/firmware.h> 40 #include <asm/page.h> 41 #include <asm/pgtable.h> 42 #include <asm/mmu.h> 43 #include <asm/mmu_context.h> 44 #include <asm/tlbflush.h> 45 #include <asm/siginfo.h> 46 #include <asm/debug.h> 47 48 #include "icswx.h" 49 50 #ifdef CONFIG_KPROBES 51 static inline int notify_page_fault(struct pt_regs *regs) 52 { 53 int ret = 0; 54 55 /* kprobe_running() needs smp_processor_id() */ 56 if (!user_mode(regs)) { 57 preempt_disable(); 58 if (kprobe_running() && kprobe_fault_handler(regs, 11)) 59 ret = 1; 60 preempt_enable(); 61 } 62 63 return ret; 64 } 65 #else 66 static inline int notify_page_fault(struct pt_regs *regs) 67 { 68 return 0; 69 } 70 #endif 71 72 /* 73 * Check whether the instruction at regs->nip is a store using 74 * an update addressing form which will update r1. 75 */ 76 static int store_updates_sp(struct pt_regs *regs) 77 { 78 unsigned int inst; 79 80 if (get_user(inst, (unsigned int __user *)regs->nip)) 81 return 0; 82 /* check for 1 in the rA field */ 83 if (((inst >> 16) & 0x1f) != 1) 84 return 0; 85 /* check major opcode */ 86 switch (inst >> 26) { 87 case 37: /* stwu */ 88 case 39: /* stbu */ 89 case 45: /* sthu */ 90 case 53: /* stfsu */ 91 case 55: /* stfdu */ 92 return 1; 93 case 62: /* std or stdu */ 94 return (inst & 3) == 1; 95 case 31: 96 /* check minor opcode */ 97 switch ((inst >> 1) & 0x3ff) { 98 case 181: /* stdux */ 99 case 183: /* stwux */ 100 case 247: /* stbux */ 101 case 439: /* sthux */ 102 case 695: /* stfsux */ 103 case 759: /* stfdux */ 104 return 1; 105 } 106 } 107 return 0; 108 } 109 /* 110 * do_page_fault error handling helpers 111 */ 112 113 #define MM_FAULT_RETURN 0 114 #define MM_FAULT_CONTINUE -1 115 #define MM_FAULT_ERR(sig) (sig) 116 117 static int do_sigbus(struct pt_regs *regs, unsigned long address, 118 unsigned int fault) 119 { 120 siginfo_t info; 121 unsigned int lsb = 0; 122 123 up_read(¤t->mm->mmap_sem); 124 125 if (!user_mode(regs)) 126 return MM_FAULT_ERR(SIGBUS); 127 128 current->thread.trap_nr = BUS_ADRERR; 129 info.si_signo = SIGBUS; 130 info.si_errno = 0; 131 info.si_code = BUS_ADRERR; 132 info.si_addr = (void __user *)address; 133 #ifdef CONFIG_MEMORY_FAILURE 134 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 135 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 136 current->comm, current->pid, address); 137 info.si_code = BUS_MCEERR_AR; 138 } 139 140 if (fault & VM_FAULT_HWPOISON_LARGE) 141 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 142 if (fault & VM_FAULT_HWPOISON) 143 lsb = PAGE_SHIFT; 144 #endif 145 info.si_addr_lsb = lsb; 146 force_sig_info(SIGBUS, &info, current); 147 return MM_FAULT_RETURN; 148 } 149 150 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault) 151 { 152 /* 153 * Pagefault was interrupted by SIGKILL. We have no reason to 154 * continue the pagefault. 155 */ 156 if (fatal_signal_pending(current)) { 157 /* 158 * If we have retry set, the mmap semaphore will have 159 * alrady been released in __lock_page_or_retry(). Else 160 * we release it now. 161 */ 162 if (!(fault & VM_FAULT_RETRY)) 163 up_read(¤t->mm->mmap_sem); 164 /* Coming from kernel, we need to deal with uaccess fixups */ 165 if (user_mode(regs)) 166 return MM_FAULT_RETURN; 167 return MM_FAULT_ERR(SIGKILL); 168 } 169 170 /* No fault: be happy */ 171 if (!(fault & VM_FAULT_ERROR)) 172 return MM_FAULT_CONTINUE; 173 174 /* Out of memory */ 175 if (fault & VM_FAULT_OOM) { 176 up_read(¤t->mm->mmap_sem); 177 178 /* 179 * We ran out of memory, or some other thing happened to us that 180 * made us unable to handle the page fault gracefully. 181 */ 182 if (!user_mode(regs)) 183 return MM_FAULT_ERR(SIGKILL); 184 pagefault_out_of_memory(); 185 return MM_FAULT_RETURN; 186 } 187 188 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) 189 return do_sigbus(regs, addr, fault); 190 191 /* We don't understand the fault code, this is fatal */ 192 BUG(); 193 return MM_FAULT_CONTINUE; 194 } 195 196 /* 197 * For 600- and 800-family processors, the error_code parameter is DSISR 198 * for a data fault, SRR1 for an instruction fault. For 400-family processors 199 * the error_code parameter is ESR for a data fault, 0 for an instruction 200 * fault. 201 * For 64-bit processors, the error_code parameter is 202 * - DSISR for a non-SLB data access fault, 203 * - SRR1 & 0x08000000 for a non-SLB instruction access fault 204 * - 0 any SLB fault. 205 * 206 * The return value is 0 if the fault was handled, or the signal 207 * number if this is a kernel fault that can't be handled here. 208 */ 209 int do_page_fault(struct pt_regs *regs, unsigned long address, 210 unsigned long error_code) 211 { 212 enum ctx_state prev_state = exception_enter(); 213 struct vm_area_struct * vma; 214 struct mm_struct *mm = current->mm; 215 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 216 int code = SEGV_MAPERR; 217 int is_write = 0; 218 int trap = TRAP(regs); 219 int is_exec = trap == 0x400; 220 int fault; 221 int rc = 0, store_update_sp = 0; 222 223 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) 224 /* 225 * Fortunately the bit assignments in SRR1 for an instruction 226 * fault and DSISR for a data fault are mostly the same for the 227 * bits we are interested in. But there are some bits which 228 * indicate errors in DSISR but can validly be set in SRR1. 229 */ 230 if (trap == 0x400) 231 error_code &= 0x48200000; 232 else 233 is_write = error_code & DSISR_ISSTORE; 234 #else 235 is_write = error_code & ESR_DST; 236 #endif /* CONFIG_4xx || CONFIG_BOOKE */ 237 238 #ifdef CONFIG_PPC_ICSWX 239 /* 240 * we need to do this early because this "data storage 241 * interrupt" does not update the DAR/DEAR so we don't want to 242 * look at it 243 */ 244 if (error_code & ICSWX_DSI_UCT) { 245 rc = acop_handle_fault(regs, address, error_code); 246 if (rc) 247 goto bail; 248 } 249 #endif /* CONFIG_PPC_ICSWX */ 250 251 if (notify_page_fault(regs)) 252 goto bail; 253 254 if (unlikely(debugger_fault_handler(regs))) 255 goto bail; 256 257 /* 258 * The kernel should never take an execute fault nor should it 259 * take a page fault to a kernel address. 260 */ 261 if (!user_mode(regs) && (is_exec || (address >= TASK_SIZE))) { 262 rc = SIGSEGV; 263 goto bail; 264 } 265 266 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \ 267 defined(CONFIG_PPC_BOOK3S_64)) 268 if (error_code & DSISR_DABRMATCH) { 269 /* breakpoint match */ 270 do_break(regs, address, error_code); 271 goto bail; 272 } 273 #endif 274 275 /* We restore the interrupt state now */ 276 if (!arch_irq_disabled_regs(regs)) 277 local_irq_enable(); 278 279 if (faulthandler_disabled() || mm == NULL) { 280 if (!user_mode(regs)) { 281 rc = SIGSEGV; 282 goto bail; 283 } 284 /* faulthandler_disabled() in user mode is really bad, 285 as is current->mm == NULL. */ 286 printk(KERN_EMERG "Page fault in user mode with " 287 "faulthandler_disabled() = %d mm = %p\n", 288 faulthandler_disabled(), mm); 289 printk(KERN_EMERG "NIP = %lx MSR = %lx\n", 290 regs->nip, regs->msr); 291 die("Weird page fault", regs, SIGSEGV); 292 } 293 294 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 295 296 /* 297 * We want to do this outside mmap_sem, because reading code around nip 298 * can result in fault, which will cause a deadlock when called with 299 * mmap_sem held 300 */ 301 if (user_mode(regs)) 302 store_update_sp = store_updates_sp(regs); 303 304 if (user_mode(regs)) 305 flags |= FAULT_FLAG_USER; 306 307 /* When running in the kernel we expect faults to occur only to 308 * addresses in user space. All other faults represent errors in the 309 * kernel and should generate an OOPS. Unfortunately, in the case of an 310 * erroneous fault occurring in a code path which already holds mmap_sem 311 * we will deadlock attempting to validate the fault against the 312 * address space. Luckily the kernel only validly references user 313 * space from well defined areas of code, which are listed in the 314 * exceptions table. 315 * 316 * As the vast majority of faults will be valid we will only perform 317 * the source reference check when there is a possibility of a deadlock. 318 * Attempt to lock the address space, if we cannot we then validate the 319 * source. If this is invalid we can skip the address space check, 320 * thus avoiding the deadlock. 321 */ 322 if (!down_read_trylock(&mm->mmap_sem)) { 323 if (!user_mode(regs) && !search_exception_tables(regs->nip)) 324 goto bad_area_nosemaphore; 325 326 retry: 327 down_read(&mm->mmap_sem); 328 } else { 329 /* 330 * The above down_read_trylock() might have succeeded in 331 * which case we'll have missed the might_sleep() from 332 * down_read(): 333 */ 334 might_sleep(); 335 } 336 337 vma = find_vma(mm, address); 338 if (!vma) 339 goto bad_area; 340 if (vma->vm_start <= address) 341 goto good_area; 342 if (!(vma->vm_flags & VM_GROWSDOWN)) 343 goto bad_area; 344 345 /* 346 * N.B. The POWER/Open ABI allows programs to access up to 347 * 288 bytes below the stack pointer. 348 * The kernel signal delivery code writes up to about 1.5kB 349 * below the stack pointer (r1) before decrementing it. 350 * The exec code can write slightly over 640kB to the stack 351 * before setting the user r1. Thus we allow the stack to 352 * expand to 1MB without further checks. 353 */ 354 if (address + 0x100000 < vma->vm_end) { 355 /* get user regs even if this fault is in kernel mode */ 356 struct pt_regs *uregs = current->thread.regs; 357 if (uregs == NULL) 358 goto bad_area; 359 360 /* 361 * A user-mode access to an address a long way below 362 * the stack pointer is only valid if the instruction 363 * is one which would update the stack pointer to the 364 * address accessed if the instruction completed, 365 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb 366 * (or the byte, halfword, float or double forms). 367 * 368 * If we don't check this then any write to the area 369 * between the last mapped region and the stack will 370 * expand the stack rather than segfaulting. 371 */ 372 if (address + 2048 < uregs->gpr[1] && !store_update_sp) 373 goto bad_area; 374 } 375 if (expand_stack(vma, address)) 376 goto bad_area; 377 378 good_area: 379 code = SEGV_ACCERR; 380 #if defined(CONFIG_6xx) 381 if (error_code & 0x95700000) 382 /* an error such as lwarx to I/O controller space, 383 address matching DABR, eciwx, etc. */ 384 goto bad_area; 385 #endif /* CONFIG_6xx */ 386 #if defined(CONFIG_8xx) 387 /* The MPC8xx seems to always set 0x80000000, which is 388 * "undefined". Of those that can be set, this is the only 389 * one which seems bad. 390 */ 391 if (error_code & 0x10000000) 392 /* Guarded storage error. */ 393 goto bad_area; 394 #endif /* CONFIG_8xx */ 395 396 if (is_exec) { 397 /* 398 * Allow execution from readable areas if the MMU does not 399 * provide separate controls over reading and executing. 400 * 401 * Note: That code used to not be enabled for 4xx/BookE. 402 * It is now as I/D cache coherency for these is done at 403 * set_pte_at() time and I see no reason why the test 404 * below wouldn't be valid on those processors. This -may- 405 * break programs compiled with a really old ABI though. 406 */ 407 if (!(vma->vm_flags & VM_EXEC) && 408 (cpu_has_feature(CPU_FTR_NOEXECUTE) || 409 !(vma->vm_flags & (VM_READ | VM_WRITE)))) 410 goto bad_area; 411 /* a write */ 412 } else if (is_write) { 413 if (!(vma->vm_flags & VM_WRITE)) 414 goto bad_area; 415 flags |= FAULT_FLAG_WRITE; 416 /* a read */ 417 } else { 418 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) 419 goto bad_area; 420 } 421 #ifdef CONFIG_PPC_STD_MMU 422 /* 423 * For hash translation mode, we should never get a 424 * PROTFAULT. Any update to pte to reduce access will result in us 425 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE 426 * fault instead of DSISR_PROTFAULT. 427 * 428 * A pte update to relax the access will not result in a hash page table 429 * entry invalidate and hence can result in DSISR_PROTFAULT. 430 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have 431 * the special !is_write in the below conditional. 432 * 433 * For platforms that doesn't supports coherent icache and do support 434 * per page noexec bit, we do setup things such that we do the 435 * sync between D/I cache via fault. But that is handled via low level 436 * hash fault code (hash_page_do_lazy_icache()) and we should not reach 437 * here in such case. 438 * 439 * For wrong access that can result in PROTFAULT, the above vma->vm_flags 440 * check should handle those and hence we should fall to the bad_area 441 * handling correctly. 442 * 443 * For embedded with per page exec support that doesn't support coherent 444 * icache we do get PROTFAULT and we handle that D/I cache sync in 445 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON 446 * is conditional for server MMU. 447 * 448 * For radix, we can get prot fault for autonuma case, because radix 449 * page table will have them marked noaccess for user. 450 */ 451 if (!radix_enabled() && !is_write) 452 WARN_ON_ONCE(error_code & DSISR_PROTFAULT); 453 #endif /* CONFIG_PPC_STD_MMU */ 454 455 /* 456 * If for any reason at all we couldn't handle the fault, 457 * make sure we exit gracefully rather than endlessly redo 458 * the fault. 459 */ 460 fault = handle_mm_fault(vma, address, flags); 461 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) { 462 if (fault & VM_FAULT_SIGSEGV) 463 goto bad_area; 464 rc = mm_fault_error(regs, address, fault); 465 if (rc >= MM_FAULT_RETURN) 466 goto bail; 467 else 468 rc = 0; 469 } 470 471 /* 472 * Major/minor page fault accounting is only done on the 473 * initial attempt. If we go through a retry, it is extremely 474 * likely that the page will be found in page cache at that point. 475 */ 476 if (flags & FAULT_FLAG_ALLOW_RETRY) { 477 if (fault & VM_FAULT_MAJOR) { 478 current->maj_flt++; 479 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 480 regs, address); 481 #ifdef CONFIG_PPC_SMLPAR 482 if (firmware_has_feature(FW_FEATURE_CMO)) { 483 u32 page_ins; 484 485 preempt_disable(); 486 page_ins = be32_to_cpu(get_lppaca()->page_ins); 487 page_ins += 1 << PAGE_FACTOR; 488 get_lppaca()->page_ins = cpu_to_be32(page_ins); 489 preempt_enable(); 490 } 491 #endif /* CONFIG_PPC_SMLPAR */ 492 } else { 493 current->min_flt++; 494 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 495 regs, address); 496 } 497 if (fault & VM_FAULT_RETRY) { 498 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk 499 * of starvation. */ 500 flags &= ~FAULT_FLAG_ALLOW_RETRY; 501 flags |= FAULT_FLAG_TRIED; 502 goto retry; 503 } 504 } 505 506 up_read(&mm->mmap_sem); 507 goto bail; 508 509 bad_area: 510 up_read(&mm->mmap_sem); 511 512 bad_area_nosemaphore: 513 /* User mode accesses cause a SIGSEGV */ 514 if (user_mode(regs)) { 515 _exception(SIGSEGV, regs, code, address); 516 goto bail; 517 } 518 519 if (is_exec && (error_code & DSISR_PROTFAULT)) 520 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected" 521 " page (%lx) - exploit attempt? (uid: %d)\n", 522 address, from_kuid(&init_user_ns, current_uid())); 523 524 rc = SIGSEGV; 525 526 bail: 527 exception_exit(prev_state); 528 return rc; 529 } 530 NOKPROBE_SYMBOL(do_page_fault); 531 532 /* 533 * bad_page_fault is called when we have a bad access from the kernel. 534 * It is called from the DSI and ISI handlers in head.S and from some 535 * of the procedures in traps.c. 536 */ 537 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) 538 { 539 const struct exception_table_entry *entry; 540 541 /* Are we prepared to handle this fault? */ 542 if ((entry = search_exception_tables(regs->nip)) != NULL) { 543 regs->nip = extable_fixup(entry); 544 return; 545 } 546 547 /* kernel has accessed a bad area */ 548 549 switch (regs->trap) { 550 case 0x300: 551 case 0x380: 552 printk(KERN_ALERT "Unable to handle kernel paging request for " 553 "data at address 0x%08lx\n", regs->dar); 554 break; 555 case 0x400: 556 case 0x480: 557 printk(KERN_ALERT "Unable to handle kernel paging request for " 558 "instruction fetch\n"); 559 break; 560 case 0x600: 561 printk(KERN_ALERT "Unable to handle kernel paging request for " 562 "unaligned access at address 0x%08lx\n", regs->dar); 563 break; 564 default: 565 printk(KERN_ALERT "Unable to handle kernel paging request for " 566 "unknown fault\n"); 567 break; 568 } 569 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", 570 regs->nip); 571 572 if (task_stack_end_corrupted(current)) 573 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); 574 575 die("Kernel access of bad area", regs, sig); 576 } 577