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/kernel.h> 21 #include <linux/errno.h> 22 #include <linux/string.h> 23 #include <linux/types.h> 24 #include <linux/ptrace.h> 25 #include <linux/mman.h> 26 #include <linux/mm.h> 27 #include <linux/interrupt.h> 28 #include <linux/highmem.h> 29 #include <linux/module.h> 30 #include <linux/kprobes.h> 31 #include <linux/kdebug.h> 32 #include <linux/perf_event.h> 33 #include <linux/magic.h> 34 #include <linux/ratelimit.h> 35 36 #include <asm/firmware.h> 37 #include <asm/page.h> 38 #include <asm/pgtable.h> 39 #include <asm/mmu.h> 40 #include <asm/mmu_context.h> 41 #include <asm/uaccess.h> 42 #include <asm/tlbflush.h> 43 #include <asm/siginfo.h> 44 #include <asm/debug.h> 45 #include <mm/mmu_decl.h> 46 47 #include "icswx.h" 48 49 #ifdef CONFIG_KPROBES 50 static inline int notify_page_fault(struct pt_regs *regs) 51 { 52 int ret = 0; 53 54 /* kprobe_running() needs smp_processor_id() */ 55 if (!user_mode(regs)) { 56 preempt_disable(); 57 if (kprobe_running() && kprobe_fault_handler(regs, 11)) 58 ret = 1; 59 preempt_enable(); 60 } 61 62 return ret; 63 } 64 #else 65 static inline int notify_page_fault(struct pt_regs *regs) 66 { 67 return 0; 68 } 69 #endif 70 71 /* 72 * Check whether the instruction at regs->nip is a store using 73 * an update addressing form which will update r1. 74 */ 75 static int store_updates_sp(struct pt_regs *regs) 76 { 77 unsigned int inst; 78 79 if (get_user(inst, (unsigned int __user *)regs->nip)) 80 return 0; 81 /* check for 1 in the rA field */ 82 if (((inst >> 16) & 0x1f) != 1) 83 return 0; 84 /* check major opcode */ 85 switch (inst >> 26) { 86 case 37: /* stwu */ 87 case 39: /* stbu */ 88 case 45: /* sthu */ 89 case 53: /* stfsu */ 90 case 55: /* stfdu */ 91 return 1; 92 case 62: /* std or stdu */ 93 return (inst & 3) == 1; 94 case 31: 95 /* check minor opcode */ 96 switch ((inst >> 1) & 0x3ff) { 97 case 181: /* stdux */ 98 case 183: /* stwux */ 99 case 247: /* stbux */ 100 case 439: /* sthux */ 101 case 695: /* stfsux */ 102 case 759: /* stfdux */ 103 return 1; 104 } 105 } 106 return 0; 107 } 108 /* 109 * do_page_fault error handling helpers 110 */ 111 112 #define MM_FAULT_RETURN 0 113 #define MM_FAULT_CONTINUE -1 114 #define MM_FAULT_ERR(sig) (sig) 115 116 static int out_of_memory(struct pt_regs *regs) 117 { 118 /* 119 * We ran out of memory, or some other thing happened to us that made 120 * us unable to handle the page fault gracefully. 121 */ 122 up_read(¤t->mm->mmap_sem); 123 if (!user_mode(regs)) 124 return MM_FAULT_ERR(SIGKILL); 125 pagefault_out_of_memory(); 126 return MM_FAULT_RETURN; 127 } 128 129 static int do_sigbus(struct pt_regs *regs, unsigned long address) 130 { 131 siginfo_t info; 132 133 up_read(¤t->mm->mmap_sem); 134 135 if (user_mode(regs)) { 136 current->thread.trap_nr = BUS_ADRERR; 137 info.si_signo = SIGBUS; 138 info.si_errno = 0; 139 info.si_code = BUS_ADRERR; 140 info.si_addr = (void __user *)address; 141 force_sig_info(SIGBUS, &info, current); 142 return MM_FAULT_RETURN; 143 } 144 return MM_FAULT_ERR(SIGBUS); 145 } 146 147 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault) 148 { 149 /* 150 * Pagefault was interrupted by SIGKILL. We have no reason to 151 * continue the pagefault. 152 */ 153 if (fatal_signal_pending(current)) { 154 /* 155 * If we have retry set, the mmap semaphore will have 156 * alrady been released in __lock_page_or_retry(). Else 157 * we release it now. 158 */ 159 if (!(fault & VM_FAULT_RETRY)) 160 up_read(¤t->mm->mmap_sem); 161 /* Coming from kernel, we need to deal with uaccess fixups */ 162 if (user_mode(regs)) 163 return MM_FAULT_RETURN; 164 return MM_FAULT_ERR(SIGKILL); 165 } 166 167 /* No fault: be happy */ 168 if (!(fault & VM_FAULT_ERROR)) 169 return MM_FAULT_CONTINUE; 170 171 /* Out of memory */ 172 if (fault & VM_FAULT_OOM) 173 return out_of_memory(regs); 174 175 /* Bus error. x86 handles HWPOISON here, we'll add this if/when 176 * we support the feature in HW 177 */ 178 if (fault & VM_FAULT_SIGBUS) 179 return do_sigbus(regs, addr); 180 181 /* We don't understand the fault code, this is fatal */ 182 BUG(); 183 return MM_FAULT_CONTINUE; 184 } 185 186 /* 187 * For 600- and 800-family processors, the error_code parameter is DSISR 188 * for a data fault, SRR1 for an instruction fault. For 400-family processors 189 * the error_code parameter is ESR for a data fault, 0 for an instruction 190 * fault. 191 * For 64-bit processors, the error_code parameter is 192 * - DSISR for a non-SLB data access fault, 193 * - SRR1 & 0x08000000 for a non-SLB instruction access fault 194 * - 0 any SLB fault. 195 * 196 * The return value is 0 if the fault was handled, or the signal 197 * number if this is a kernel fault that can't be handled here. 198 */ 199 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address, 200 unsigned long error_code) 201 { 202 struct vm_area_struct * vma; 203 struct mm_struct *mm = current->mm; 204 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 205 int code = SEGV_MAPERR; 206 int is_write = 0; 207 int trap = TRAP(regs); 208 int is_exec = trap == 0x400; 209 int fault; 210 211 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) 212 /* 213 * Fortunately the bit assignments in SRR1 for an instruction 214 * fault and DSISR for a data fault are mostly the same for the 215 * bits we are interested in. But there are some bits which 216 * indicate errors in DSISR but can validly be set in SRR1. 217 */ 218 if (trap == 0x400) 219 error_code &= 0x48200000; 220 else 221 is_write = error_code & DSISR_ISSTORE; 222 #else 223 is_write = error_code & ESR_DST; 224 #endif /* CONFIG_4xx || CONFIG_BOOKE */ 225 226 if (is_write) 227 flags |= FAULT_FLAG_WRITE; 228 229 #ifdef CONFIG_PPC_ICSWX 230 /* 231 * we need to do this early because this "data storage 232 * interrupt" does not update the DAR/DEAR so we don't want to 233 * look at it 234 */ 235 if (error_code & ICSWX_DSI_UCT) { 236 int rc = acop_handle_fault(regs, address, error_code); 237 if (rc) 238 return rc; 239 } 240 #endif /* CONFIG_PPC_ICSWX */ 241 242 if (notify_page_fault(regs)) 243 return 0; 244 245 if (unlikely(debugger_fault_handler(regs))) 246 return 0; 247 248 /* On a kernel SLB miss we can only check for a valid exception entry */ 249 if (!user_mode(regs) && (address >= TASK_SIZE)) 250 return SIGSEGV; 251 252 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \ 253 defined(CONFIG_PPC_BOOK3S_64)) 254 if (error_code & DSISR_DABRMATCH) { 255 /* DABR match */ 256 do_dabr(regs, address, error_code); 257 return 0; 258 } 259 #endif 260 261 /* We restore the interrupt state now */ 262 if (!arch_irq_disabled_regs(regs)) 263 local_irq_enable(); 264 265 if (in_atomic() || mm == NULL) { 266 if (!user_mode(regs)) 267 return SIGSEGV; 268 /* in_atomic() in user mode is really bad, 269 as is current->mm == NULL. */ 270 printk(KERN_EMERG "Page fault in user mode with " 271 "in_atomic() = %d mm = %p\n", in_atomic(), mm); 272 printk(KERN_EMERG "NIP = %lx MSR = %lx\n", 273 regs->nip, regs->msr); 274 die("Weird page fault", regs, SIGSEGV); 275 } 276 277 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 278 279 /* When running in the kernel we expect faults to occur only to 280 * addresses in user space. All other faults represent errors in the 281 * kernel and should generate an OOPS. Unfortunately, in the case of an 282 * erroneous fault occurring in a code path which already holds mmap_sem 283 * we will deadlock attempting to validate the fault against the 284 * address space. Luckily the kernel only validly references user 285 * space from well defined areas of code, which are listed in the 286 * exceptions table. 287 * 288 * As the vast majority of faults will be valid we will only perform 289 * the source reference check when there is a possibility of a deadlock. 290 * Attempt to lock the address space, if we cannot we then validate the 291 * source. If this is invalid we can skip the address space check, 292 * thus avoiding the deadlock. 293 */ 294 if (!down_read_trylock(&mm->mmap_sem)) { 295 if (!user_mode(regs) && !search_exception_tables(regs->nip)) 296 goto bad_area_nosemaphore; 297 298 retry: 299 down_read(&mm->mmap_sem); 300 } else { 301 /* 302 * The above down_read_trylock() might have succeeded in 303 * which case we'll have missed the might_sleep() from 304 * down_read(): 305 */ 306 might_sleep(); 307 } 308 309 vma = find_vma(mm, address); 310 if (!vma) 311 goto bad_area; 312 if (vma->vm_start <= address) 313 goto good_area; 314 if (!(vma->vm_flags & VM_GROWSDOWN)) 315 goto bad_area; 316 317 /* 318 * N.B. The POWER/Open ABI allows programs to access up to 319 * 288 bytes below the stack pointer. 320 * The kernel signal delivery code writes up to about 1.5kB 321 * below the stack pointer (r1) before decrementing it. 322 * The exec code can write slightly over 640kB to the stack 323 * before setting the user r1. Thus we allow the stack to 324 * expand to 1MB without further checks. 325 */ 326 if (address + 0x100000 < vma->vm_end) { 327 /* get user regs even if this fault is in kernel mode */ 328 struct pt_regs *uregs = current->thread.regs; 329 if (uregs == NULL) 330 goto bad_area; 331 332 /* 333 * A user-mode access to an address a long way below 334 * the stack pointer is only valid if the instruction 335 * is one which would update the stack pointer to the 336 * address accessed if the instruction completed, 337 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb 338 * (or the byte, halfword, float or double forms). 339 * 340 * If we don't check this then any write to the area 341 * between the last mapped region and the stack will 342 * expand the stack rather than segfaulting. 343 */ 344 if (address + 2048 < uregs->gpr[1] 345 && (!user_mode(regs) || !store_updates_sp(regs))) 346 goto bad_area; 347 } 348 if (expand_stack(vma, address)) 349 goto bad_area; 350 351 good_area: 352 code = SEGV_ACCERR; 353 #if defined(CONFIG_6xx) 354 if (error_code & 0x95700000) 355 /* an error such as lwarx to I/O controller space, 356 address matching DABR, eciwx, etc. */ 357 goto bad_area; 358 #endif /* CONFIG_6xx */ 359 #if defined(CONFIG_8xx) 360 /* 8xx sometimes need to load a invalid/non-present TLBs. 361 * These must be invalidated separately as linux mm don't. 362 */ 363 if (error_code & 0x40000000) /* no translation? */ 364 _tlbil_va(address, 0, 0, 0); 365 366 /* The MPC8xx seems to always set 0x80000000, which is 367 * "undefined". Of those that can be set, this is the only 368 * one which seems bad. 369 */ 370 if (error_code & 0x10000000) 371 /* Guarded storage error. */ 372 goto bad_area; 373 #endif /* CONFIG_8xx */ 374 375 if (is_exec) { 376 #ifdef CONFIG_PPC_STD_MMU 377 /* Protection fault on exec go straight to failure on 378 * Hash based MMUs as they either don't support per-page 379 * execute permission, or if they do, it's handled already 380 * at the hash level. This test would probably have to 381 * be removed if we change the way this works to make hash 382 * processors use the same I/D cache coherency mechanism 383 * as embedded. 384 */ 385 if (error_code & DSISR_PROTFAULT) 386 goto bad_area; 387 #endif /* CONFIG_PPC_STD_MMU */ 388 389 /* 390 * Allow execution from readable areas if the MMU does not 391 * provide separate controls over reading and executing. 392 * 393 * Note: That code used to not be enabled for 4xx/BookE. 394 * It is now as I/D cache coherency for these is done at 395 * set_pte_at() time and I see no reason why the test 396 * below wouldn't be valid on those processors. This -may- 397 * break programs compiled with a really old ABI though. 398 */ 399 if (!(vma->vm_flags & VM_EXEC) && 400 (cpu_has_feature(CPU_FTR_NOEXECUTE) || 401 !(vma->vm_flags & (VM_READ | VM_WRITE)))) 402 goto bad_area; 403 /* a write */ 404 } else if (is_write) { 405 if (!(vma->vm_flags & VM_WRITE)) 406 goto bad_area; 407 /* a read */ 408 } else { 409 /* protection fault */ 410 if (error_code & 0x08000000) 411 goto bad_area; 412 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) 413 goto bad_area; 414 } 415 416 /* 417 * If for any reason at all we couldn't handle the fault, 418 * make sure we exit gracefully rather than endlessly redo 419 * the fault. 420 */ 421 fault = handle_mm_fault(mm, vma, address, flags); 422 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) { 423 int rc = mm_fault_error(regs, address, fault); 424 if (rc >= MM_FAULT_RETURN) 425 return rc; 426 } 427 428 /* 429 * Major/minor page fault accounting is only done on the 430 * initial attempt. If we go through a retry, it is extremely 431 * likely that the page will be found in page cache at that point. 432 */ 433 if (flags & FAULT_FLAG_ALLOW_RETRY) { 434 if (fault & VM_FAULT_MAJOR) { 435 current->maj_flt++; 436 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 437 regs, address); 438 #ifdef CONFIG_PPC_SMLPAR 439 if (firmware_has_feature(FW_FEATURE_CMO)) { 440 preempt_disable(); 441 get_lppaca()->page_ins += (1 << PAGE_FACTOR); 442 preempt_enable(); 443 } 444 #endif /* CONFIG_PPC_SMLPAR */ 445 } else { 446 current->min_flt++; 447 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 448 regs, address); 449 } 450 if (fault & VM_FAULT_RETRY) { 451 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk 452 * of starvation. */ 453 flags &= ~FAULT_FLAG_ALLOW_RETRY; 454 flags |= FAULT_FLAG_TRIED; 455 goto retry; 456 } 457 } 458 459 up_read(&mm->mmap_sem); 460 return 0; 461 462 bad_area: 463 up_read(&mm->mmap_sem); 464 465 bad_area_nosemaphore: 466 /* User mode accesses cause a SIGSEGV */ 467 if (user_mode(regs)) { 468 _exception(SIGSEGV, regs, code, address); 469 return 0; 470 } 471 472 if (is_exec && (error_code & DSISR_PROTFAULT)) 473 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected" 474 " page (%lx) - exploit attempt? (uid: %d)\n", 475 address, from_kuid(&init_user_ns, current_uid())); 476 477 return SIGSEGV; 478 479 } 480 481 /* 482 * bad_page_fault is called when we have a bad access from the kernel. 483 * It is called from the DSI and ISI handlers in head.S and from some 484 * of the procedures in traps.c. 485 */ 486 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) 487 { 488 const struct exception_table_entry *entry; 489 unsigned long *stackend; 490 491 /* Are we prepared to handle this fault? */ 492 if ((entry = search_exception_tables(regs->nip)) != NULL) { 493 regs->nip = entry->fixup; 494 return; 495 } 496 497 /* kernel has accessed a bad area */ 498 499 switch (regs->trap) { 500 case 0x300: 501 case 0x380: 502 printk(KERN_ALERT "Unable to handle kernel paging request for " 503 "data at address 0x%08lx\n", regs->dar); 504 break; 505 case 0x400: 506 case 0x480: 507 printk(KERN_ALERT "Unable to handle kernel paging request for " 508 "instruction fetch\n"); 509 break; 510 default: 511 printk(KERN_ALERT "Unable to handle kernel paging request for " 512 "unknown fault\n"); 513 break; 514 } 515 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", 516 regs->nip); 517 518 stackend = end_of_stack(current); 519 if (current != &init_task && *stackend != STACK_END_MAGIC) 520 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); 521 522 die("Kernel access of bad area", regs, sig); 523 } 524