Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/i386/i386/trap.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (C) 1994, David Greenman 3 * Copyright (c) 1990, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * the University of Utah, and William Jolitz. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the University of 20 * California, Berkeley and its contributors. 21 * 4. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD: releng/11.1/sys/i386/i386/trap.c 333371 2018-05-08 17:12:10Z gordon $"); 42 43 /* 44 * 386 Trap and System call handling 45 */ 46 47 #include "opt_clock.h" 48 #include "opt_cpu.h" 49 #include "opt_hwpmc_hooks.h" 50 #include "opt_isa.h" 51 #include "opt_kdb.h" 52 #include "opt_stack.h" 53 #include "opt_trap.h" 54 55 #include <sys/param.h> 56 #include <sys/bus.h> 57 #include <sys/systm.h> 58 #include <sys/proc.h> 59 #include <sys/pioctl.h> 60 #include <sys/ptrace.h> 61 #include <sys/kdb.h> 62 #include <sys/kernel.h> 63 #include <sys/ktr.h> 64 #include <sys/lock.h> 65 #include <sys/mutex.h> 66 #include <sys/resourcevar.h> 67 #include <sys/signalvar.h> 68 #include <sys/syscall.h> 69 #include <sys/sysctl.h> 70 #include <sys/sysent.h> 71 #include <sys/uio.h> 72 #include <sys/vmmeter.h> 73 #ifdef HWPMC_HOOKS 74 #include <sys/pmckern.h> 75 PMC_SOFT_DEFINE( , , page_fault, all); 76 PMC_SOFT_DEFINE( , , page_fault, read); 77 PMC_SOFT_DEFINE( , , page_fault, write); 78 #endif 79 #include <security/audit/audit.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_param.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_kern.h> 85 #include <vm/vm_map.h> 86 #include <vm/vm_page.h> 87 #include <vm/vm_extern.h> 88 89 #include <machine/cpu.h> 90 #include <machine/intr_machdep.h> 91 #include <x86/mca.h> 92 #include <machine/md_var.h> 93 #include <machine/pcb.h> 94 #ifdef SMP 95 #include <machine/smp.h> 96 #endif 97 #include <machine/stack.h> 98 #include <machine/tss.h> 99 #include <machine/vm86.h> 100 101 #ifdef POWERFAIL_NMI 102 #include <sys/syslog.h> 103 #include <machine/clock.h> 104 #endif 105 106 #ifdef KDTRACE_HOOKS 107 #include <sys/dtrace_bsd.h> 108 #endif 109 110 extern void trap(struct trapframe *frame); 111 extern void syscall(struct trapframe *frame); 112 113 static int trap_pfault(struct trapframe *, int, vm_offset_t); 114 static void trap_fatal(struct trapframe *, vm_offset_t); 115 void dblfault_handler(void); 116 117 extern inthand_t IDTVEC(lcall_syscall); 118 119 extern inthand_t IDTVEC(bpt), IDTVEC(dbg), IDTVEC(int0x80_syscall); 120 121 #define MAX_TRAP_MSG 32 122 static char *trap_msg[] = { 123 "", /* 0 unused */ 124 "privileged instruction fault", /* 1 T_PRIVINFLT */ 125 "", /* 2 unused */ 126 "breakpoint instruction fault", /* 3 T_BPTFLT */ 127 "", /* 4 unused */ 128 "", /* 5 unused */ 129 "arithmetic trap", /* 6 T_ARITHTRAP */ 130 "", /* 7 unused */ 131 "", /* 8 unused */ 132 "general protection fault", /* 9 T_PROTFLT */ 133 "trace trap", /* 10 T_TRCTRAP */ 134 "", /* 11 unused */ 135 "page fault", /* 12 T_PAGEFLT */ 136 "", /* 13 unused */ 137 "alignment fault", /* 14 T_ALIGNFLT */ 138 "", /* 15 unused */ 139 "", /* 16 unused */ 140 "", /* 17 unused */ 141 "integer divide fault", /* 18 T_DIVIDE */ 142 "non-maskable interrupt trap", /* 19 T_NMI */ 143 "overflow trap", /* 20 T_OFLOW */ 144 "FPU bounds check fault", /* 21 T_BOUND */ 145 "FPU device not available", /* 22 T_DNA */ 146 "double fault", /* 23 T_DOUBLEFLT */ 147 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 148 "invalid TSS fault", /* 25 T_TSSFLT */ 149 "segment not present fault", /* 26 T_SEGNPFLT */ 150 "stack fault", /* 27 T_STKFLT */ 151 "machine check trap", /* 28 T_MCHK */ 152 "SIMD floating-point exception", /* 29 T_XMMFLT */ 153 "reserved (unknown) fault", /* 30 T_RESERVED */ 154 "", /* 31 unused (reserved) */ 155 "DTrace pid return trap", /* 32 T_DTRACE_RET */ 156 }; 157 158 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 159 int has_f00f_bug = 0; /* Initialized so that it can be patched. */ 160 #endif 161 162 static int prot_fault_translation = 0; 163 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW, 164 &prot_fault_translation, 0, "Select signal to deliver on protection fault"); 165 static int uprintf_signal; 166 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW, 167 &uprintf_signal, 0, 168 "Print debugging information on trap signal to ctty"); 169 170 /* 171 * Exception, fault, and trap interface to the FreeBSD kernel. 172 * This common code is called from assembly language IDT gate entry 173 * routines that prepare a suitable stack frame, and restore this 174 * frame after the exception has been processed. 175 */ 176 177 void 178 trap(struct trapframe *frame) 179 { 180 #ifdef KDTRACE_HOOKS 181 struct reg regs; 182 #endif 183 struct thread *td = curthread; 184 struct proc *p = td->td_proc; 185 #ifdef KDB 186 register_t dr6; 187 #endif 188 int i = 0, ucode = 0; 189 u_int type; 190 register_t addr = 0; 191 vm_offset_t eva; 192 ksiginfo_t ksi; 193 #ifdef POWERFAIL_NMI 194 static int lastalert = 0; 195 #endif 196 197 PCPU_INC(cnt.v_trap); 198 type = frame->tf_trapno; 199 200 #ifdef SMP 201 /* Handler for NMI IPIs used for stopping CPUs. */ 202 if (type == T_NMI) { 203 if (ipi_nmi_handler() == 0) 204 goto out; 205 } 206 #endif /* SMP */ 207 208 #ifdef KDB 209 if (kdb_active) { 210 kdb_reenter(); 211 goto out; 212 } 213 #endif 214 215 if (type == T_RESERVED) { 216 trap_fatal(frame, 0); 217 goto out; 218 } 219 220 if (type == T_NMI) { 221 #ifdef HWPMC_HOOKS 222 /* 223 * CPU PMCs interrupt using an NMI so we check for that first. 224 * If the HWPMC module is active, 'pmc_hook' will point to 225 * the function to be called. A non-zero return value from the 226 * hook means that the NMI was consumed by it and that we can 227 * return immediately. 228 */ 229 if (pmc_intr != NULL && 230 (*pmc_intr)(PCPU_GET(cpuid), frame) != 0) 231 goto out; 232 #endif 233 234 #ifdef STACK 235 if (stack_nmi_handler(frame) != 0) 236 goto out; 237 #endif 238 } 239 240 if (type == T_MCHK) { 241 mca_intr(); 242 goto out; 243 } 244 245 #ifdef KDTRACE_HOOKS 246 /* 247 * A trap can occur while DTrace executes a probe. Before 248 * executing the probe, DTrace blocks re-scheduling and sets 249 * a flag in its per-cpu flags to indicate that it doesn't 250 * want to fault. On returning from the probe, the no-fault 251 * flag is cleared and finally re-scheduling is enabled. 252 */ 253 if ((type == T_PROTFLT || type == T_PAGEFLT) && 254 dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type)) 255 goto out; 256 #endif 257 258 if ((frame->tf_eflags & PSL_I) == 0) { 259 /* 260 * Buggy application or kernel code has disabled 261 * interrupts and then trapped. Enabling interrupts 262 * now is wrong, but it is better than running with 263 * interrupts disabled until they are accidentally 264 * enabled later. 265 */ 266 if (TRAPF_USERMODE(frame) && 267 (curpcb->pcb_flags & PCB_VM86CALL) == 0) 268 uprintf( 269 "pid %ld (%s): trap %d with interrupts disabled\n", 270 (long)curproc->p_pid, curthread->td_name, type); 271 else if (type != T_NMI && type != T_BPTFLT && 272 type != T_TRCTRAP && 273 frame->tf_eip != (int)cpu_switch_load_gs) { 274 /* 275 * XXX not quite right, since this may be for a 276 * multiple fault in user mode. 277 */ 278 printf("kernel trap %d with interrupts disabled\n", 279 type); 280 /* 281 * Page faults need interrupts disabled until later, 282 * and we shouldn't enable interrupts while holding 283 * a spin lock. 284 */ 285 if (type != T_PAGEFLT && 286 td->td_md.md_spinlock_count == 0) 287 enable_intr(); 288 } 289 } 290 eva = 0; 291 if (type == T_PAGEFLT) { 292 /* 293 * For some Cyrix CPUs, %cr2 is clobbered by 294 * interrupts. This problem is worked around by using 295 * an interrupt gate for the pagefault handler. We 296 * are finally ready to read %cr2 and conditionally 297 * reenable interrupts. If we hold a spin lock, then 298 * we must not reenable interrupts. This might be a 299 * spurious page fault. 300 */ 301 eva = rcr2(); 302 if (td->td_md.md_spinlock_count == 0) 303 enable_intr(); 304 } 305 306 if (TRAPF_USERMODE(frame) && (curpcb->pcb_flags & PCB_VM86CALL) == 0) { 307 /* user trap */ 308 309 td->td_pticks = 0; 310 td->td_frame = frame; 311 addr = frame->tf_eip; 312 if (td->td_cowgen != p->p_cowgen) 313 thread_cow_update(td); 314 315 switch (type) { 316 case T_PRIVINFLT: /* privileged instruction fault */ 317 i = SIGILL; 318 ucode = ILL_PRVOPC; 319 break; 320 321 case T_BPTFLT: /* bpt instruction fault */ 322 case T_TRCTRAP: /* trace trap */ 323 enable_intr(); 324 #ifdef KDTRACE_HOOKS 325 if (type == T_BPTFLT) { 326 fill_frame_regs(frame, ®s); 327 if (dtrace_pid_probe_ptr != NULL && 328 dtrace_pid_probe_ptr(®s) == 0) 329 goto out; 330 } 331 #endif 332 user_trctrap_out: 333 frame->tf_eflags &= ~PSL_T; 334 i = SIGTRAP; 335 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 336 break; 337 338 case T_ARITHTRAP: /* arithmetic trap */ 339 ucode = npxtrap_x87(); 340 if (ucode == -1) 341 goto userout; 342 i = SIGFPE; 343 break; 344 345 /* 346 * The following two traps can happen in 347 * vm86 mode, and, if so, we want to handle 348 * them specially. 349 */ 350 case T_PROTFLT: /* general protection fault */ 351 case T_STKFLT: /* stack fault */ 352 if (frame->tf_eflags & PSL_VM) { 353 i = vm86_emulate((struct vm86frame *)frame); 354 if (i == SIGTRAP) { 355 type = T_TRCTRAP; 356 load_dr6(rdr6() | 0x4000); 357 goto user_trctrap_out; 358 } 359 if (i == 0) 360 goto user; 361 break; 362 } 363 i = SIGBUS; 364 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR; 365 break; 366 case T_SEGNPFLT: /* segment not present fault */ 367 i = SIGBUS; 368 ucode = BUS_ADRERR; 369 break; 370 case T_TSSFLT: /* invalid TSS fault */ 371 i = SIGBUS; 372 ucode = BUS_OBJERR; 373 break; 374 case T_ALIGNFLT: 375 i = SIGBUS; 376 ucode = BUS_ADRALN; 377 break; 378 case T_DOUBLEFLT: /* double fault */ 379 default: 380 i = SIGBUS; 381 ucode = BUS_OBJERR; 382 break; 383 384 case T_PAGEFLT: /* page fault */ 385 386 i = trap_pfault(frame, TRUE, eva); 387 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 388 if (i == -2) { 389 /* 390 * The f00f hack workaround has triggered, so 391 * treat the fault as an illegal instruction 392 * (T_PRIVINFLT) instead of a page fault. 393 */ 394 type = frame->tf_trapno = T_PRIVINFLT; 395 396 /* Proceed as in that case. */ 397 ucode = ILL_PRVOPC; 398 i = SIGILL; 399 break; 400 } 401 #endif 402 if (i == -1) 403 goto userout; 404 if (i == 0) 405 goto user; 406 407 if (i == SIGSEGV) 408 ucode = SEGV_MAPERR; 409 else { 410 if (prot_fault_translation == 0) { 411 /* 412 * Autodetect. 413 * This check also covers the images 414 * without the ABI-tag ELF note. 415 */ 416 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD 417 && p->p_osrel >= P_OSREL_SIGSEGV) { 418 i = SIGSEGV; 419 ucode = SEGV_ACCERR; 420 } else { 421 i = SIGBUS; 422 ucode = BUS_PAGE_FAULT; 423 } 424 } else if (prot_fault_translation == 1) { 425 /* 426 * Always compat mode. 427 */ 428 i = SIGBUS; 429 ucode = BUS_PAGE_FAULT; 430 } else { 431 /* 432 * Always SIGSEGV mode. 433 */ 434 i = SIGSEGV; 435 ucode = SEGV_ACCERR; 436 } 437 } 438 addr = eva; 439 break; 440 441 case T_DIVIDE: /* integer divide fault */ 442 ucode = FPE_INTDIV; 443 i = SIGFPE; 444 break; 445 446 #ifdef DEV_ISA 447 case T_NMI: 448 #ifdef POWERFAIL_NMI 449 #ifndef TIMER_FREQ 450 # define TIMER_FREQ 1193182 451 #endif 452 if (time_second - lastalert > 10) { 453 log(LOG_WARNING, "NMI: power fail\n"); 454 sysbeep(880, hz); 455 lastalert = time_second; 456 } 457 goto userout; 458 #else /* !POWERFAIL_NMI */ 459 nmi_handle_intr(type, frame); 460 break; 461 #endif /* POWERFAIL_NMI */ 462 #endif /* DEV_ISA */ 463 464 case T_OFLOW: /* integer overflow fault */ 465 ucode = FPE_INTOVF; 466 i = SIGFPE; 467 break; 468 469 case T_BOUND: /* bounds check fault */ 470 ucode = FPE_FLTSUB; 471 i = SIGFPE; 472 break; 473 474 case T_DNA: 475 KASSERT(PCB_USER_FPU(td->td_pcb), 476 ("kernel FPU ctx has leaked")); 477 /* transparent fault (due to context switch "late") */ 478 if (npxdna()) 479 goto userout; 480 uprintf("pid %d killed due to lack of floating point\n", 481 p->p_pid); 482 i = SIGKILL; 483 ucode = 0; 484 break; 485 486 case T_FPOPFLT: /* FPU operand fetch fault */ 487 ucode = ILL_COPROC; 488 i = SIGILL; 489 break; 490 491 case T_XMMFLT: /* SIMD floating-point exception */ 492 ucode = npxtrap_sse(); 493 if (ucode == -1) 494 goto userout; 495 i = SIGFPE; 496 break; 497 #ifdef KDTRACE_HOOKS 498 case T_DTRACE_RET: 499 enable_intr(); 500 fill_frame_regs(frame, ®s); 501 if (dtrace_return_probe_ptr != NULL && 502 dtrace_return_probe_ptr(®s) == 0) 503 goto out; 504 break; 505 #endif 506 } 507 } else { 508 /* kernel trap */ 509 510 KASSERT(cold || td->td_ucred != NULL, 511 ("kernel trap doesn't have ucred")); 512 switch (type) { 513 case T_PAGEFLT: /* page fault */ 514 (void) trap_pfault(frame, FALSE, eva); 515 goto out; 516 517 case T_DNA: 518 if (PCB_USER_FPU(td->td_pcb)) 519 panic("Unregistered use of FPU in kernel"); 520 if (npxdna()) 521 goto out; 522 break; 523 524 case T_ARITHTRAP: /* arithmetic trap */ 525 case T_XMMFLT: /* SIMD floating-point exception */ 526 case T_FPOPFLT: /* FPU operand fetch fault */ 527 /* 528 * XXXKIB for now disable any FPU traps in kernel 529 * handler registration seems to be overkill 530 */ 531 trap_fatal(frame, 0); 532 goto out; 533 534 /* 535 * The following two traps can happen in 536 * vm86 mode, and, if so, we want to handle 537 * them specially. 538 */ 539 case T_PROTFLT: /* general protection fault */ 540 case T_STKFLT: /* stack fault */ 541 if (frame->tf_eflags & PSL_VM) { 542 i = vm86_emulate((struct vm86frame *)frame); 543 if (i == SIGTRAP) { 544 type = T_TRCTRAP; 545 load_dr6(rdr6() | 0x4000); 546 goto kernel_trctrap; 547 } 548 if (i != 0) 549 /* 550 * returns to original process 551 */ 552 vm86_trap((struct vm86frame *)frame); 553 goto out; 554 } 555 if (type == T_STKFLT) 556 break; 557 558 /* FALL THROUGH */ 559 560 case T_SEGNPFLT: /* segment not present fault */ 561 if (curpcb->pcb_flags & PCB_VM86CALL) 562 break; 563 564 /* 565 * Invalid %fs's and %gs's can be created using 566 * procfs or PT_SETREGS or by invalidating the 567 * underlying LDT entry. This causes a fault 568 * in kernel mode when the kernel attempts to 569 * switch contexts. Lose the bad context 570 * (XXX) so that we can continue, and generate 571 * a signal. 572 */ 573 if (frame->tf_eip == (int)cpu_switch_load_gs) { 574 curpcb->pcb_gs = 0; 575 #if 0 576 PROC_LOCK(p); 577 kern_psignal(p, SIGBUS); 578 PROC_UNLOCK(p); 579 #endif 580 goto out; 581 } 582 583 if (td->td_intr_nesting_level != 0) 584 break; 585 586 /* 587 * Invalid segment selectors and out of bounds 588 * %eip's and %esp's can be set up in user mode. 589 * This causes a fault in kernel mode when the 590 * kernel tries to return to user mode. We want 591 * to get this fault so that we can fix the 592 * problem here and not have to check all the 593 * selectors and pointers when the user changes 594 * them. 595 */ 596 if (frame->tf_eip == (int)doreti_iret) { 597 frame->tf_eip = (int)doreti_iret_fault; 598 goto out; 599 } 600 if (frame->tf_eip == (int)doreti_popl_ds) { 601 frame->tf_eip = (int)doreti_popl_ds_fault; 602 goto out; 603 } 604 if (frame->tf_eip == (int)doreti_popl_es) { 605 frame->tf_eip = (int)doreti_popl_es_fault; 606 goto out; 607 } 608 if (frame->tf_eip == (int)doreti_popl_fs) { 609 frame->tf_eip = (int)doreti_popl_fs_fault; 610 goto out; 611 } 612 if (curpcb->pcb_onfault != NULL) { 613 frame->tf_eip = 614 (int)curpcb->pcb_onfault; 615 goto out; 616 } 617 break; 618 619 case T_TSSFLT: 620 /* 621 * PSL_NT can be set in user mode and isn't cleared 622 * automatically when the kernel is entered. This 623 * causes a TSS fault when the kernel attempts to 624 * `iret' because the TSS link is uninitialized. We 625 * want to get this fault so that we can fix the 626 * problem here and not every time the kernel is 627 * entered. 628 */ 629 if (frame->tf_eflags & PSL_NT) { 630 frame->tf_eflags &= ~PSL_NT; 631 goto out; 632 } 633 break; 634 635 case T_TRCTRAP: /* trace trap */ 636 kernel_trctrap: 637 if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) { 638 /* 639 * We've just entered system mode via the 640 * syscall lcall. Continue single stepping 641 * silently until the syscall handler has 642 * saved the flags. 643 */ 644 goto out; 645 } 646 if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) { 647 /* 648 * The syscall handler has now saved the 649 * flags. Stop single stepping it. 650 */ 651 frame->tf_eflags &= ~PSL_T; 652 goto out; 653 } 654 /* 655 * Ignore debug register trace traps due to 656 * accesses in the user's address space, which 657 * can happen under several conditions such as 658 * if a user sets a watchpoint on a buffer and 659 * then passes that buffer to a system call. 660 * We still want to get TRCTRAPS for addresses 661 * in kernel space because that is useful when 662 * debugging the kernel. 663 */ 664 if (user_dbreg_trap() && 665 !(curpcb->pcb_flags & PCB_VM86CALL)) { 666 /* 667 * Reset breakpoint bits because the 668 * processor doesn't 669 */ 670 load_dr6(rdr6() & ~0xf); 671 goto out; 672 } 673 674 /* 675 * Malicious user code can configure a debug 676 * register watchpoint to trap on data access 677 * to the top of stack and then execute 'pop 678 * %ss; int 3'. Due to exception deferral for 679 * 'pop %ss', the CPU will not interrupt 'int 680 * 3' to raise the DB# exception for the debug 681 * register but will postpone the DB# until 682 * execution of the first instruction of the 683 * BP# handler (in kernel mode). Normally the 684 * previous check would ignore DB# exceptions 685 * for watchpoints on user addresses raised in 686 * kernel mode. However, some CPU errata 687 * include cases where DB# exceptions do not 688 * properly set bits in %dr6, e.g. Haswell 689 * HSD23 and Skylake-X SKZ24. 690 * 691 * A deferred DB# can also be raised on the 692 * first instructions of system call entry 693 * points or single-step traps via similar use 694 * of 'pop %ss' or 'mov xxx, %ss'. 695 */ 696 if (frame->tf_eip == 697 (uintptr_t)IDTVEC(int0x80_syscall) || 698 frame->tf_eip == (uintptr_t)IDTVEC(bpt) || 699 frame->tf_eip == (uintptr_t)IDTVEC(dbg)) 700 return; 701 /* 702 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 703 */ 704 case T_BPTFLT: 705 /* 706 * If KDB is enabled, let it handle the debugger trap. 707 * Otherwise, debugger traps "can't happen". 708 */ 709 #ifdef KDB 710 /* XXX %dr6 is not quite reentrant. */ 711 dr6 = rdr6(); 712 load_dr6(dr6 & ~0x4000); 713 if (kdb_trap(type, dr6, frame)) 714 goto out; 715 #endif 716 break; 717 718 #ifdef DEV_ISA 719 case T_NMI: 720 #ifdef POWERFAIL_NMI 721 if (time_second - lastalert > 10) { 722 log(LOG_WARNING, "NMI: power fail\n"); 723 sysbeep(880, hz); 724 lastalert = time_second; 725 } 726 goto out; 727 #else /* !POWERFAIL_NMI */ 728 nmi_handle_intr(type, frame); 729 goto out; 730 #endif /* POWERFAIL_NMI */ 731 #endif /* DEV_ISA */ 732 } 733 734 trap_fatal(frame, eva); 735 goto out; 736 } 737 738 /* Translate fault for emulators (e.g. Linux) */ 739 if (*p->p_sysent->sv_transtrap) 740 i = (*p->p_sysent->sv_transtrap)(i, type); 741 742 ksiginfo_init_trap(&ksi); 743 ksi.ksi_signo = i; 744 ksi.ksi_code = ucode; 745 ksi.ksi_addr = (void *)addr; 746 ksi.ksi_trapno = type; 747 if (uprintf_signal) { 748 uprintf("pid %d comm %s: signal %d err %x code %d type %d " 749 "addr 0x%x esp 0x%08x eip 0x%08x " 750 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n", 751 p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr, 752 frame->tf_esp, frame->tf_eip, 753 fubyte((void *)(frame->tf_eip + 0)), 754 fubyte((void *)(frame->tf_eip + 1)), 755 fubyte((void *)(frame->tf_eip + 2)), 756 fubyte((void *)(frame->tf_eip + 3)), 757 fubyte((void *)(frame->tf_eip + 4)), 758 fubyte((void *)(frame->tf_eip + 5)), 759 fubyte((void *)(frame->tf_eip + 6)), 760 fubyte((void *)(frame->tf_eip + 7))); 761 } 762 KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled")); 763 trapsignal(td, &ksi); 764 765 #ifdef DEBUG 766 if (type <= MAX_TRAP_MSG) { 767 uprintf("fatal process exception: %s", 768 trap_msg[type]); 769 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 770 uprintf(", fault VA = 0x%lx", (u_long)eva); 771 uprintf("\n"); 772 } 773 #endif 774 775 user: 776 userret(td, frame); 777 KASSERT(PCB_USER_FPU(td->td_pcb), 778 ("Return from trap with kernel FPU ctx leaked")); 779 userout: 780 out: 781 return; 782 } 783 784 static int 785 trap_pfault(frame, usermode, eva) 786 struct trapframe *frame; 787 int usermode; 788 vm_offset_t eva; 789 { 790 vm_offset_t va; 791 vm_map_t map; 792 int rv = 0; 793 vm_prot_t ftype; 794 struct thread *td = curthread; 795 struct proc *p = td->td_proc; 796 797 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) { 798 /* 799 * Due to both processor errata and lazy TLB invalidation when 800 * access restrictions are removed from virtual pages, memory 801 * accesses that are allowed by the physical mapping layer may 802 * nonetheless cause one spurious page fault per virtual page. 803 * When the thread is executing a "no faulting" section that 804 * is bracketed by vm_fault_{disable,enable}_pagefaults(), 805 * every page fault is treated as a spurious page fault, 806 * unless it accesses the same virtual address as the most 807 * recent page fault within the same "no faulting" section. 808 */ 809 if (td->td_md.md_spurflt_addr != eva || 810 (td->td_pflags & TDP_RESETSPUR) != 0) { 811 /* 812 * Do nothing to the TLB. A stale TLB entry is 813 * flushed automatically by a page fault. 814 */ 815 td->td_md.md_spurflt_addr = eva; 816 td->td_pflags &= ~TDP_RESETSPUR; 817 return (0); 818 } 819 } else { 820 /* 821 * If we get a page fault while in a critical section, then 822 * it is most likely a fatal kernel page fault. The kernel 823 * is already going to panic trying to get a sleep lock to 824 * do the VM lookup, so just consider it a fatal trap so the 825 * kernel can print out a useful trap message and even get 826 * to the debugger. 827 * 828 * If we get a page fault while holding a non-sleepable 829 * lock, then it is most likely a fatal kernel page fault. 830 * If WITNESS is enabled, then it's going to whine about 831 * bogus LORs with various VM locks, so just skip to the 832 * fatal trap handling directly. 833 */ 834 if (td->td_critnest != 0 || 835 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL, 836 "Kernel page fault") != 0) { 837 trap_fatal(frame, eva); 838 return (-1); 839 } 840 } 841 va = trunc_page(eva); 842 if (va >= KERNBASE) { 843 /* 844 * Don't allow user-mode faults in kernel address space. 845 * An exception: if the faulting address is the invalid 846 * instruction entry in the IDT, then the Intel Pentium 847 * F00F bug workaround was triggered, and we need to 848 * treat it is as an illegal instruction, and not a page 849 * fault. 850 */ 851 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 852 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) 853 return (-2); 854 #endif 855 if (usermode) 856 goto nogo; 857 858 map = kernel_map; 859 } else { 860 map = &p->p_vmspace->vm_map; 861 862 /* 863 * When accessing a user-space address, kernel must be 864 * ready to accept the page fault, and provide a 865 * handling routine. Since accessing the address 866 * without the handler is a bug, do not try to handle 867 * it normally, and panic immediately. 868 */ 869 if (!usermode && (td->td_intr_nesting_level != 0 || 870 curpcb->pcb_onfault == NULL)) { 871 trap_fatal(frame, eva); 872 return (-1); 873 } 874 } 875 876 /* 877 * If the trap was caused by errant bits in the PTE then panic. 878 */ 879 if (frame->tf_err & PGEX_RSV) { 880 trap_fatal(frame, eva); 881 return (-1); 882 } 883 884 /* 885 * PGEX_I is defined only if the execute disable bit capability is 886 * supported and enabled. 887 */ 888 if (frame->tf_err & PGEX_W) 889 ftype = VM_PROT_WRITE; 890 #if defined(PAE) || defined(PAE_TABLES) 891 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 892 ftype = VM_PROT_EXECUTE; 893 #endif 894 else 895 ftype = VM_PROT_READ; 896 897 /* Fault in the page. */ 898 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 899 if (rv == KERN_SUCCESS) { 900 #ifdef HWPMC_HOOKS 901 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 902 PMC_SOFT_CALL_TF( , , page_fault, all, frame); 903 if (ftype == VM_PROT_READ) 904 PMC_SOFT_CALL_TF( , , page_fault, read, 905 frame); 906 else 907 PMC_SOFT_CALL_TF( , , page_fault, write, 908 frame); 909 } 910 #endif 911 return (0); 912 } 913 nogo: 914 if (!usermode) { 915 if (td->td_intr_nesting_level == 0 && 916 curpcb->pcb_onfault != NULL) { 917 frame->tf_eip = (int)curpcb->pcb_onfault; 918 return (0); 919 } 920 trap_fatal(frame, eva); 921 return (-1); 922 } 923 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 924 } 925 926 static void 927 trap_fatal(frame, eva) 928 struct trapframe *frame; 929 vm_offset_t eva; 930 { 931 int code, ss, esp; 932 u_int type; 933 struct soft_segment_descriptor softseg; 934 char *msg; 935 936 code = frame->tf_err; 937 type = frame->tf_trapno; 938 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 939 940 if (type <= MAX_TRAP_MSG) 941 msg = trap_msg[type]; 942 else 943 msg = "UNKNOWN"; 944 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 945 frame->tf_eflags & PSL_VM ? "vm86" : 946 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 947 #ifdef SMP 948 /* two separate prints in case of a trap on an unmapped page */ 949 printf("cpuid = %d; ", PCPU_GET(cpuid)); 950 printf("apic id = %02x\n", PCPU_GET(apic_id)); 951 #endif 952 if (type == T_PAGEFLT) { 953 printf("fault virtual address = 0x%x\n", eva); 954 printf("fault code = %s %s%s, %s\n", 955 code & PGEX_U ? "user" : "supervisor", 956 code & PGEX_W ? "write" : "read", 957 #if defined(PAE) || defined(PAE_TABLES) 958 pg_nx != 0 ? 959 (code & PGEX_I ? " instruction" : " data") : 960 #endif 961 "", 962 code & PGEX_RSV ? "reserved bits in PTE" : 963 code & PGEX_P ? "protection violation" : "page not present"); 964 } 965 printf("instruction pointer = 0x%x:0x%x\n", 966 frame->tf_cs & 0xffff, frame->tf_eip); 967 if (TF_HAS_STACKREGS(frame)) { 968 ss = frame->tf_ss & 0xffff; 969 esp = frame->tf_esp; 970 } else { 971 ss = GSEL(GDATA_SEL, SEL_KPL); 972 esp = (int)&frame->tf_esp; 973 } 974 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 975 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 976 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 977 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 978 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 979 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 980 softseg.ssd_gran); 981 printf("processor eflags = "); 982 if (frame->tf_eflags & PSL_T) 983 printf("trace trap, "); 984 if (frame->tf_eflags & PSL_I) 985 printf("interrupt enabled, "); 986 if (frame->tf_eflags & PSL_NT) 987 printf("nested task, "); 988 if (frame->tf_eflags & PSL_RF) 989 printf("resume, "); 990 if (frame->tf_eflags & PSL_VM) 991 printf("vm86, "); 992 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 993 printf("current process = %d (%s)\n", 994 curproc->p_pid, curthread->td_name); 995 996 #ifdef KDB 997 if (debugger_on_panic || kdb_active) { 998 frame->tf_err = eva; /* smuggle fault address to ddb */ 999 if (kdb_trap(type, 0, frame)) { 1000 frame->tf_err = code; /* restore error code */ 1001 return; 1002 } 1003 frame->tf_err = code; /* restore error code */ 1004 } 1005 #endif 1006 printf("trap number = %d\n", type); 1007 if (type <= MAX_TRAP_MSG) 1008 panic("%s", trap_msg[type]); 1009 else 1010 panic("unknown/reserved trap"); 1011 } 1012 1013 /* 1014 * Double fault handler. Called when a fault occurs while writing 1015 * a frame for a trap/exception onto the stack. This usually occurs 1016 * when the stack overflows (such is the case with infinite recursion, 1017 * for example). 1018 * 1019 * XXX Note that the current PTD gets replaced by IdlePTD when the 1020 * task switch occurs. This means that the stack that was active at 1021 * the time of the double fault is not available at <kstack> unless 1022 * the machine was idle when the double fault occurred. The downside 1023 * of this is that "trace <ebp>" in ddb won't work. 1024 */ 1025 void 1026 dblfault_handler() 1027 { 1028 #ifdef KDTRACE_HOOKS 1029 if (dtrace_doubletrap_func != NULL) 1030 (*dtrace_doubletrap_func)(); 1031 #endif 1032 printf("\nFatal double fault:\n"); 1033 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip)); 1034 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp)); 1035 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp)); 1036 #ifdef SMP 1037 /* two separate prints in case of a trap on an unmapped page */ 1038 printf("cpuid = %d; ", PCPU_GET(cpuid)); 1039 printf("apic id = %02x\n", PCPU_GET(apic_id)); 1040 #endif 1041 panic("double fault"); 1042 } 1043 1044 int 1045 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa) 1046 { 1047 struct proc *p; 1048 struct trapframe *frame; 1049 caddr_t params; 1050 long tmp; 1051 int error; 1052 1053 p = td->td_proc; 1054 frame = td->td_frame; 1055 1056 params = (caddr_t)frame->tf_esp + sizeof(int); 1057 sa->code = frame->tf_eax; 1058 1059 /* 1060 * Need to check if this is a 32 bit or 64 bit syscall. 1061 */ 1062 if (sa->code == SYS_syscall) { 1063 /* 1064 * Code is first argument, followed by actual args. 1065 */ 1066 error = fueword(params, &tmp); 1067 if (error == -1) 1068 return (EFAULT); 1069 sa->code = tmp; 1070 params += sizeof(int); 1071 } else if (sa->code == SYS___syscall) { 1072 /* 1073 * Like syscall, but code is a quad, so as to maintain 1074 * quad alignment for the rest of the arguments. 1075 */ 1076 error = fueword(params, &tmp); 1077 if (error == -1) 1078 return (EFAULT); 1079 sa->code = tmp; 1080 params += sizeof(quad_t); 1081 } 1082 1083 if (p->p_sysent->sv_mask) 1084 sa->code &= p->p_sysent->sv_mask; 1085 if (sa->code >= p->p_sysent->sv_size) 1086 sa->callp = &p->p_sysent->sv_table[0]; 1087 else 1088 sa->callp = &p->p_sysent->sv_table[sa->code]; 1089 sa->narg = sa->callp->sy_narg; 1090 1091 if (params != NULL && sa->narg != 0) 1092 error = copyin(params, (caddr_t)sa->args, 1093 (u_int)(sa->narg * sizeof(int))); 1094 else 1095 error = 0; 1096 1097 if (error == 0) { 1098 td->td_retval[0] = 0; 1099 td->td_retval[1] = frame->tf_edx; 1100 } 1101 1102 return (error); 1103 } 1104 1105 #include "../../kern/subr_syscall.c" 1106 1107 /* 1108 * syscall - system call request C handler. A system call is 1109 * essentially treated as a trap by reusing the frame layout. 1110 */ 1111 void 1112 syscall(struct trapframe *frame) 1113 { 1114 struct thread *td; 1115 struct syscall_args sa; 1116 register_t orig_tf_eflags; 1117 int error; 1118 ksiginfo_t ksi; 1119 1120 #ifdef DIAGNOSTIC 1121 if (!(TRAPF_USERMODE(frame) && 1122 (curpcb->pcb_flags & PCB_VM86CALL) == 0)) { 1123 panic("syscall"); 1124 /* NOT REACHED */ 1125 } 1126 #endif 1127 orig_tf_eflags = frame->tf_eflags; 1128 1129 td = curthread; 1130 td->td_frame = frame; 1131 1132 error = syscallenter(td, &sa); 1133 1134 /* 1135 * Traced syscall. 1136 */ 1137 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) { 1138 frame->tf_eflags &= ~PSL_T; 1139 ksiginfo_init_trap(&ksi); 1140 ksi.ksi_signo = SIGTRAP; 1141 ksi.ksi_code = TRAP_TRACE; 1142 ksi.ksi_addr = (void *)frame->tf_eip; 1143 trapsignal(td, &ksi); 1144 } 1145 1146 KASSERT(PCB_USER_FPU(td->td_pcb), 1147 ("System call %s returning with kernel FPU ctx leaked", 1148 syscallname(td->td_proc, sa.code))); 1149 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td), 1150 ("System call %s returning with mangled pcb_save", 1151 syscallname(td->td_proc, sa.code))); 1152 1153 syscallret(td, error, &sa); 1154 }
Cache object: 1fc89c5368ccc7eea0eaf607228b6e70
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]