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("$FreeBSD$"); 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_kdtrace.h" 53 #include "opt_ktrace.h" 54 #include "opt_npx.h" 55 #include "opt_trap.h" 56 57 #include <sys/param.h> 58 #include <sys/bus.h> 59 #include <sys/systm.h> 60 #include <sys/proc.h> 61 #include <sys/pioctl.h> 62 #include <sys/ptrace.h> 63 #include <sys/kdb.h> 64 #include <sys/kernel.h> 65 #include <sys/ktr.h> 66 #include <sys/lock.h> 67 #include <sys/mutex.h> 68 #include <sys/resourcevar.h> 69 #include <sys/signalvar.h> 70 #include <sys/syscall.h> 71 #include <sys/sysctl.h> 72 #include <sys/sysent.h> 73 #include <sys/uio.h> 74 #include <sys/vmmeter.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 #ifdef HWPMC_HOOKS 79 #include <sys/pmckern.h> 80 #endif 81 #include <security/audit/audit.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_param.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_kern.h> 87 #include <vm/vm_map.h> 88 #include <vm/vm_page.h> 89 #include <vm/vm_extern.h> 90 91 #include <machine/cpu.h> 92 #include <machine/intr_machdep.h> 93 #include <machine/mca.h> 94 #include <machine/md_var.h> 95 #include <machine/pcb.h> 96 #ifdef SMP 97 #include <machine/smp.h> 98 #endif 99 #include <machine/tss.h> 100 #include <machine/vm86.h> 101 102 #ifdef POWERFAIL_NMI 103 #include <sys/syslog.h> 104 #include <machine/clock.h> 105 #endif 106 107 #ifdef KDTRACE_HOOKS 108 #include <sys/dtrace_bsd.h> 109 110 /* 111 * This is a hook which is initialised by the dtrace module 112 * to handle traps which might occur during DTrace probe 113 * execution. 114 */ 115 dtrace_trap_func_t dtrace_trap_func; 116 117 dtrace_doubletrap_func_t dtrace_doubletrap_func; 118 119 /* 120 * This is a hook which is initialised by the systrace module 121 * when it is loaded. This keeps the DTrace syscall provider 122 * implementation opaque. 123 */ 124 systrace_probe_func_t systrace_probe_func; 125 #endif 126 127 /* Defined in i386/i386/elf_machdep.c. */ 128 extern struct sysentvec elf32_freebsd_sysvec; 129 130 extern void trap(struct trapframe *frame); 131 extern void syscall(struct trapframe *frame); 132 133 static int trap_pfault(struct trapframe *, int, vm_offset_t); 134 static void trap_fatal(struct trapframe *, vm_offset_t); 135 void dblfault_handler(void); 136 137 extern inthand_t IDTVEC(lcall_syscall); 138 139 #define MAX_TRAP_MSG 30 140 static char *trap_msg[] = { 141 "", /* 0 unused */ 142 "privileged instruction fault", /* 1 T_PRIVINFLT */ 143 "", /* 2 unused */ 144 "breakpoint instruction fault", /* 3 T_BPTFLT */ 145 "", /* 4 unused */ 146 "", /* 5 unused */ 147 "arithmetic trap", /* 6 T_ARITHTRAP */ 148 "", /* 7 unused */ 149 "", /* 8 unused */ 150 "general protection fault", /* 9 T_PROTFLT */ 151 "trace trap", /* 10 T_TRCTRAP */ 152 "", /* 11 unused */ 153 "page fault", /* 12 T_PAGEFLT */ 154 "", /* 13 unused */ 155 "alignment fault", /* 14 T_ALIGNFLT */ 156 "", /* 15 unused */ 157 "", /* 16 unused */ 158 "", /* 17 unused */ 159 "integer divide fault", /* 18 T_DIVIDE */ 160 "non-maskable interrupt trap", /* 19 T_NMI */ 161 "overflow trap", /* 20 T_OFLOW */ 162 "FPU bounds check fault", /* 21 T_BOUND */ 163 "FPU device not available", /* 22 T_DNA */ 164 "double fault", /* 23 T_DOUBLEFLT */ 165 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 166 "invalid TSS fault", /* 25 T_TSSFLT */ 167 "segment not present fault", /* 26 T_SEGNPFLT */ 168 "stack fault", /* 27 T_STKFLT */ 169 "machine check trap", /* 28 T_MCHK */ 170 "SIMD floating-point exception", /* 29 T_XMMFLT */ 171 "reserved (unknown) fault", /* 30 T_RESERVED */ 172 }; 173 174 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 175 extern int has_f00f_bug; 176 #endif 177 178 #ifdef KDB 179 static int kdb_on_nmi = 1; 180 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW, 181 &kdb_on_nmi, 0, "Go to KDB on NMI"); 182 #endif 183 static int panic_on_nmi = 1; 184 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 185 &panic_on_nmi, 0, "Panic on NMI"); 186 static int prot_fault_translation = 0; 187 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW, 188 &prot_fault_translation, 0, "Select signal to deliver on protection fault"); 189 190 extern char *syscallnames[]; 191 192 /* 193 * Exception, fault, and trap interface to the FreeBSD kernel. 194 * This common code is called from assembly language IDT gate entry 195 * routines that prepare a suitable stack frame, and restore this 196 * frame after the exception has been processed. 197 */ 198 199 void 200 trap(struct trapframe *frame) 201 { 202 struct thread *td = curthread; 203 struct proc *p = td->td_proc; 204 int i = 0, ucode = 0, code; 205 u_int type; 206 register_t addr = 0; 207 vm_offset_t eva; 208 ksiginfo_t ksi; 209 #ifdef POWERFAIL_NMI 210 static int lastalert = 0; 211 #endif 212 213 PCPU_INC(cnt.v_trap); 214 type = frame->tf_trapno; 215 216 #ifdef SMP 217 #ifdef STOP_NMI 218 /* Handler for NMI IPIs used for stopping CPUs. */ 219 if (type == T_NMI) { 220 if (ipi_nmi_handler() == 0) 221 goto out; 222 } 223 #endif /* STOP_NMI */ 224 #endif /* SMP */ 225 226 #ifdef KDB 227 if (kdb_active) { 228 kdb_reenter(); 229 goto out; 230 } 231 #endif 232 233 #ifdef HWPMC_HOOKS 234 /* 235 * CPU PMCs interrupt using an NMI so we check for that first. 236 * If the HWPMC module is active, 'pmc_hook' will point to 237 * the function to be called. A return value of '1' from the 238 * hook means that the NMI was handled by it and that we can 239 * return immediately. 240 */ 241 if (type == T_NMI && pmc_intr && 242 (*pmc_intr)(PCPU_GET(cpuid), frame)) 243 goto out; 244 #endif 245 246 if (type == T_MCHK) { 247 if (!mca_intr()) 248 trap_fatal(frame, 0); 249 goto out; 250 } 251 252 #ifdef KDTRACE_HOOKS 253 /* 254 * A trap can occur while DTrace executes a probe. Before 255 * executing the probe, DTrace blocks re-scheduling and sets 256 * a flag in it's per-cpu flags to indicate that it doesn't 257 * want to fault. On returning from the probe, the no-fault 258 * flag is cleared and finally re-scheduling is enabled. 259 * 260 * If the DTrace kernel module has registered a trap handler, 261 * call it and if it returns non-zero, assume that it has 262 * handled the trap and modified the trap frame so that this 263 * function can return normally. 264 */ 265 if ((type == T_PROTFLT || type == T_PAGEFLT) && 266 dtrace_trap_func != NULL) 267 if ((*dtrace_trap_func)(frame, type)) 268 goto out; 269 #endif 270 271 if ((frame->tf_eflags & PSL_I) == 0) { 272 /* 273 * Buggy application or kernel code has disabled 274 * interrupts and then trapped. Enabling interrupts 275 * now is wrong, but it is better than running with 276 * interrupts disabled until they are accidentally 277 * enabled later. 278 */ 279 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM)) 280 printf( 281 "pid %ld (%s): trap %d with interrupts disabled\n", 282 (long)curproc->p_pid, curproc->p_comm, type); 283 else if (type != T_BPTFLT && type != T_TRCTRAP && 284 frame->tf_eip != (int)cpu_switch_load_gs) { 285 /* 286 * XXX not quite right, since this may be for a 287 * multiple fault in user mode. 288 */ 289 printf("kernel trap %d with interrupts disabled\n", 290 type); 291 /* 292 * Page faults need interrupts disabled until later, 293 * and we shouldn't enable interrupts while holding 294 * a spin lock or if servicing an NMI. 295 */ 296 if (type != T_NMI && type != T_PAGEFLT && 297 td->td_md.md_spinlock_count == 0) 298 enable_intr(); 299 } 300 } 301 302 eva = 0; 303 code = frame->tf_err; 304 if (type == T_PAGEFLT) { 305 /* 306 * For some Cyrix CPUs, %cr2 is clobbered by 307 * interrupts. This problem is worked around by using 308 * an interrupt gate for the pagefault handler. We 309 * are finally ready to read %cr2 and then must 310 * reenable interrupts. 311 * 312 * If we get a page fault while in a critical section, then 313 * it is most likely a fatal kernel page fault. The kernel 314 * is already going to panic trying to get a sleep lock to 315 * do the VM lookup, so just consider it a fatal trap so the 316 * kernel can print out a useful trap message and even get 317 * to the debugger. 318 * 319 * If we get a page fault while holding a non-sleepable 320 * lock, then it is most likely a fatal kernel page fault. 321 * If WITNESS is enabled, then it's going to whine about 322 * bogus LORs with various VM locks, so just skip to the 323 * fatal trap handling directly. 324 */ 325 eva = rcr2(); 326 if (td->td_critnest != 0 || 327 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL, 328 "Kernel page fault") != 0) 329 trap_fatal(frame, eva); 330 else 331 enable_intr(); 332 } 333 334 if ((ISPL(frame->tf_cs) == SEL_UPL) || 335 ((frame->tf_eflags & PSL_VM) && 336 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) { 337 /* user trap */ 338 339 td->td_pticks = 0; 340 td->td_frame = frame; 341 addr = frame->tf_eip; 342 if (td->td_ucred != p->p_ucred) 343 cred_update_thread(td); 344 345 switch (type) { 346 case T_PRIVINFLT: /* privileged instruction fault */ 347 i = SIGILL; 348 ucode = ILL_PRVOPC; 349 break; 350 351 case T_BPTFLT: /* bpt instruction fault */ 352 case T_TRCTRAP: /* trace trap */ 353 enable_intr(); 354 frame->tf_eflags &= ~PSL_T; 355 i = SIGTRAP; 356 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 357 break; 358 359 case T_ARITHTRAP: /* arithmetic trap */ 360 #ifdef DEV_NPX 361 ucode = npxtrap(); 362 if (ucode == -1) 363 goto userout; 364 #else 365 ucode = 0; 366 #endif 367 i = SIGFPE; 368 break; 369 370 /* 371 * The following two traps can happen in 372 * vm86 mode, and, if so, we want to handle 373 * them specially. 374 */ 375 case T_PROTFLT: /* general protection fault */ 376 case T_STKFLT: /* stack fault */ 377 if (frame->tf_eflags & PSL_VM) { 378 i = vm86_emulate((struct vm86frame *)frame); 379 if (i == 0) 380 goto user; 381 break; 382 } 383 i = SIGBUS; 384 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR; 385 break; 386 case T_SEGNPFLT: /* segment not present fault */ 387 i = SIGBUS; 388 ucode = BUS_ADRERR; 389 break; 390 case T_TSSFLT: /* invalid TSS fault */ 391 i = SIGBUS; 392 ucode = BUS_OBJERR; 393 break; 394 case T_DOUBLEFLT: /* double fault */ 395 default: 396 i = SIGBUS; 397 ucode = BUS_OBJERR; 398 break; 399 400 case T_PAGEFLT: /* page fault */ 401 #ifdef KSE 402 if (td->td_pflags & TDP_SA) 403 thread_user_enter(td); 404 #endif 405 406 i = trap_pfault(frame, TRUE, eva); 407 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 408 if (i == -2) { 409 /* 410 * The f00f hack workaround has triggered, so 411 * treat the fault as an illegal instruction 412 * (T_PRIVINFLT) instead of a page fault. 413 */ 414 type = frame->tf_trapno = T_PRIVINFLT; 415 416 /* Proceed as in that case. */ 417 ucode = ILL_PRVOPC; 418 i = SIGILL; 419 break; 420 } 421 #endif 422 if (i == -1) 423 goto userout; 424 if (i == 0) 425 goto user; 426 427 if (i == SIGSEGV) 428 ucode = SEGV_MAPERR; 429 else { 430 if (prot_fault_translation == 0) { 431 /* 432 * Autodetect. 433 * This check also covers the images 434 * without the ABI-tag ELF note. 435 */ 436 if (curproc->p_sysent == 437 &elf32_freebsd_sysvec && 438 p->p_osrel >= 700004) { 439 i = SIGSEGV; 440 ucode = SEGV_ACCERR; 441 } else { 442 i = SIGBUS; 443 ucode = BUS_PAGE_FAULT; 444 } 445 } else if (prot_fault_translation == 1) { 446 /* 447 * Always compat mode. 448 */ 449 i = SIGBUS; 450 ucode = BUS_PAGE_FAULT; 451 } else { 452 /* 453 * Always SIGSEGV mode. 454 */ 455 i = SIGSEGV; 456 ucode = SEGV_ACCERR; 457 } 458 } 459 addr = eva; 460 break; 461 462 case T_DIVIDE: /* integer divide fault */ 463 ucode = FPE_INTDIV; 464 i = SIGFPE; 465 break; 466 467 #ifdef DEV_ISA 468 case T_NMI: 469 #ifdef POWERFAIL_NMI 470 #ifndef TIMER_FREQ 471 # define TIMER_FREQ 1193182 472 #endif 473 mtx_lock(&Giant); 474 if (time_second - lastalert > 10) { 475 log(LOG_WARNING, "NMI: power fail\n"); 476 sysbeep(TIMER_FREQ/880, hz); 477 lastalert = time_second; 478 } 479 mtx_unlock(&Giant); 480 goto userout; 481 #else /* !POWERFAIL_NMI */ 482 /* machine/parity/power fail/"kitchen sink" faults */ 483 /* XXX Giant */ 484 if (isa_nmi(code) == 0) { 485 #ifdef KDB 486 /* 487 * NMI can be hooked up to a pushbutton 488 * for debugging. 489 */ 490 if (kdb_on_nmi) { 491 printf ("NMI ... going to debugger\n"); 492 kdb_trap(type, 0, frame); 493 } 494 #endif /* KDB */ 495 goto userout; 496 } else if (panic_on_nmi) 497 panic("NMI indicates hardware failure"); 498 break; 499 #endif /* POWERFAIL_NMI */ 500 #endif /* DEV_ISA */ 501 502 case T_OFLOW: /* integer overflow fault */ 503 ucode = FPE_INTOVF; 504 i = SIGFPE; 505 break; 506 507 case T_BOUND: /* bounds check fault */ 508 ucode = FPE_FLTSUB; 509 i = SIGFPE; 510 break; 511 512 case T_DNA: 513 #ifdef DEV_NPX 514 /* transparent fault (due to context switch "late") */ 515 if (npxdna()) 516 goto userout; 517 #endif 518 printf("pid %d killed due to lack of floating point\n", 519 p->p_pid); 520 i = SIGKILL; 521 ucode = 0; 522 break; 523 524 case T_FPOPFLT: /* FPU operand fetch fault */ 525 ucode = ILL_COPROC; 526 i = SIGILL; 527 break; 528 529 case T_XMMFLT: /* SIMD floating-point exception */ 530 ucode = 0; /* XXX */ 531 i = SIGFPE; 532 break; 533 } 534 } else { 535 /* kernel trap */ 536 537 KASSERT(cold || td->td_ucred != NULL, 538 ("kernel trap doesn't have ucred")); 539 switch (type) { 540 case T_PAGEFLT: /* page fault */ 541 (void) trap_pfault(frame, FALSE, eva); 542 goto out; 543 544 case T_DNA: 545 #ifdef DEV_NPX 546 /* 547 * The kernel is apparently using npx for copying. 548 * XXX this should be fatal unless the kernel has 549 * registered such use. 550 */ 551 if (npxdna()) 552 goto out; 553 #endif 554 break; 555 556 /* 557 * The following two traps can happen in 558 * vm86 mode, and, if so, we want to handle 559 * them specially. 560 */ 561 case T_PROTFLT: /* general protection fault */ 562 case T_STKFLT: /* stack fault */ 563 if (frame->tf_eflags & PSL_VM) { 564 i = vm86_emulate((struct vm86frame *)frame); 565 if (i != 0) 566 /* 567 * returns to original process 568 */ 569 vm86_trap((struct vm86frame *)frame); 570 goto out; 571 } 572 if (type == T_STKFLT) 573 break; 574 575 /* FALL THROUGH */ 576 577 case T_SEGNPFLT: /* segment not present fault */ 578 if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL) 579 break; 580 581 /* 582 * Invalid %fs's and %gs's can be created using 583 * procfs or PT_SETREGS or by invalidating the 584 * underlying LDT entry. This causes a fault 585 * in kernel mode when the kernel attempts to 586 * switch contexts. Lose the bad context 587 * (XXX) so that we can continue, and generate 588 * a signal. 589 */ 590 if (frame->tf_eip == (int)cpu_switch_load_gs) { 591 PCPU_GET(curpcb)->pcb_gs = 0; 592 #if 0 593 PROC_LOCK(p); 594 psignal(p, SIGBUS); 595 PROC_UNLOCK(p); 596 #endif 597 goto out; 598 } 599 600 if (td->td_intr_nesting_level != 0) 601 break; 602 603 /* 604 * Invalid segment selectors and out of bounds 605 * %eip's and %esp's can be set up in user mode. 606 * This causes a fault in kernel mode when the 607 * kernel tries to return to user mode. We want 608 * to get this fault so that we can fix the 609 * problem here and not have to check all the 610 * selectors and pointers when the user changes 611 * them. 612 */ 613 if (frame->tf_eip == (int)doreti_iret) { 614 frame->tf_eip = (int)doreti_iret_fault; 615 goto out; 616 } 617 if (frame->tf_eip == (int)doreti_popl_ds) { 618 frame->tf_eip = (int)doreti_popl_ds_fault; 619 goto out; 620 } 621 if (frame->tf_eip == (int)doreti_popl_es) { 622 frame->tf_eip = (int)doreti_popl_es_fault; 623 goto out; 624 } 625 if (frame->tf_eip == (int)doreti_popl_fs) { 626 frame->tf_eip = (int)doreti_popl_fs_fault; 627 goto out; 628 } 629 if (PCPU_GET(curpcb)->pcb_onfault != NULL) { 630 frame->tf_eip = 631 (int)PCPU_GET(curpcb)->pcb_onfault; 632 goto out; 633 } 634 break; 635 636 case T_TSSFLT: 637 /* 638 * PSL_NT can be set in user mode and isn't cleared 639 * automatically when the kernel is entered. This 640 * causes a TSS fault when the kernel attempts to 641 * `iret' because the TSS link is uninitialized. We 642 * want to get this fault so that we can fix the 643 * problem here and not every time the kernel is 644 * entered. 645 */ 646 if (frame->tf_eflags & PSL_NT) { 647 frame->tf_eflags &= ~PSL_NT; 648 goto out; 649 } 650 break; 651 652 case T_TRCTRAP: /* trace trap */ 653 if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) { 654 /* 655 * We've just entered system mode via the 656 * syscall lcall. Continue single stepping 657 * silently until the syscall handler has 658 * saved the flags. 659 */ 660 goto out; 661 } 662 if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) { 663 /* 664 * The syscall handler has now saved the 665 * flags. Stop single stepping it. 666 */ 667 frame->tf_eflags &= ~PSL_T; 668 goto out; 669 } 670 /* 671 * Ignore debug register trace traps due to 672 * accesses in the user's address space, which 673 * can happen under several conditions such as 674 * if a user sets a watchpoint on a buffer and 675 * then passes that buffer to a system call. 676 * We still want to get TRCTRAPS for addresses 677 * in kernel space because that is useful when 678 * debugging the kernel. 679 */ 680 /* XXX Giant */ 681 if (user_dbreg_trap() && 682 !(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) { 683 /* 684 * Reset breakpoint bits because the 685 * processor doesn't 686 */ 687 load_dr6(rdr6() & 0xfffffff0); 688 goto out; 689 } 690 /* 691 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 692 */ 693 case T_BPTFLT: 694 /* 695 * If KDB is enabled, let it handle the debugger trap. 696 * Otherwise, debugger traps "can't happen". 697 */ 698 #ifdef KDB 699 if (kdb_trap(type, 0, frame)) 700 goto out; 701 #endif 702 break; 703 704 #ifdef DEV_ISA 705 case T_NMI: 706 #ifdef POWERFAIL_NMI 707 mtx_lock(&Giant); 708 if (time_second - lastalert > 10) { 709 log(LOG_WARNING, "NMI: power fail\n"); 710 sysbeep(TIMER_FREQ/880, hz); 711 lastalert = time_second; 712 } 713 mtx_unlock(&Giant); 714 goto out; 715 #else /* !POWERFAIL_NMI */ 716 /* XXX Giant */ 717 /* machine/parity/power fail/"kitchen sink" faults */ 718 if (isa_nmi(code) == 0) { 719 #ifdef KDB 720 /* 721 * NMI can be hooked up to a pushbutton 722 * for debugging. 723 */ 724 if (kdb_on_nmi) { 725 printf ("NMI ... going to debugger\n"); 726 kdb_trap(type, 0, frame); 727 } 728 #endif /* KDB */ 729 goto out; 730 } else if (panic_on_nmi == 0) 731 goto out; 732 /* FALLTHROUGH */ 733 #endif /* POWERFAIL_NMI */ 734 #endif /* DEV_ISA */ 735 } 736 737 trap_fatal(frame, eva); 738 goto out; 739 } 740 741 /* Translate fault for emulators (e.g. Linux) */ 742 if (*p->p_sysent->sv_transtrap) 743 i = (*p->p_sysent->sv_transtrap)(i, type); 744 745 ksiginfo_init_trap(&ksi); 746 ksi.ksi_signo = i; 747 ksi.ksi_code = ucode; 748 ksi.ksi_addr = (void *)addr; 749 ksi.ksi_trapno = type; 750 trapsignal(td, &ksi); 751 752 #ifdef DEBUG 753 if (type <= MAX_TRAP_MSG) { 754 uprintf("fatal process exception: %s", 755 trap_msg[type]); 756 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 757 uprintf(", fault VA = 0x%lx", (u_long)eva); 758 uprintf("\n"); 759 } 760 #endif 761 762 user: 763 userret(td, frame); 764 mtx_assert(&Giant, MA_NOTOWNED); 765 userout: 766 out: 767 return; 768 } 769 770 static int 771 trap_pfault(frame, usermode, eva) 772 struct trapframe *frame; 773 int usermode; 774 vm_offset_t eva; 775 { 776 vm_offset_t va; 777 struct vmspace *vm = NULL; 778 vm_map_t map; 779 int rv = 0; 780 vm_prot_t ftype; 781 struct thread *td = curthread; 782 struct proc *p = td->td_proc; 783 784 va = trunc_page(eva); 785 if (va >= KERNBASE) { 786 /* 787 * Don't allow user-mode faults in kernel address space. 788 * An exception: if the faulting address is the invalid 789 * instruction entry in the IDT, then the Intel Pentium 790 * F00F bug workaround was triggered, and we need to 791 * treat it is as an illegal instruction, and not a page 792 * fault. 793 */ 794 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 795 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) 796 return -2; 797 #endif 798 if (usermode) 799 goto nogo; 800 801 map = kernel_map; 802 } else { 803 /* 804 * This is a fault on non-kernel virtual memory. 805 * vm is initialized above to NULL. If curproc is NULL 806 * or curproc->p_vmspace is NULL the fault is fatal. 807 */ 808 if (p != NULL) 809 vm = p->p_vmspace; 810 811 if (vm == NULL) 812 goto nogo; 813 814 map = &vm->vm_map; 815 } 816 817 /* 818 * PGEX_I is defined only if the execute disable bit capability is 819 * supported and enabled. 820 */ 821 if (frame->tf_err & PGEX_W) 822 ftype = VM_PROT_WRITE; 823 #ifdef PAE 824 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 825 ftype = VM_PROT_EXECUTE; 826 #endif 827 else 828 ftype = VM_PROT_READ; 829 830 if (map != kernel_map) { 831 /* 832 * Keep swapout from messing with us during this 833 * critical time. 834 */ 835 PROC_LOCK(p); 836 ++p->p_lock; 837 PROC_UNLOCK(p); 838 839 /* Fault in the user page: */ 840 rv = vm_fault(map, va, ftype, 841 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 842 : VM_FAULT_NORMAL); 843 844 PROC_LOCK(p); 845 --p->p_lock; 846 PROC_UNLOCK(p); 847 } else { 848 /* 849 * Don't have to worry about process locking or stacks in the 850 * kernel. 851 */ 852 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 853 } 854 if (rv == KERN_SUCCESS) 855 return (0); 856 nogo: 857 if (!usermode) { 858 if (td->td_intr_nesting_level == 0 && 859 PCPU_GET(curpcb)->pcb_onfault != NULL) { 860 frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault; 861 return (0); 862 } 863 trap_fatal(frame, eva); 864 return (-1); 865 } 866 867 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 868 } 869 870 static void 871 trap_fatal(frame, eva) 872 struct trapframe *frame; 873 vm_offset_t eva; 874 { 875 int code, ss, esp; 876 u_int type; 877 struct soft_segment_descriptor softseg; 878 char *msg; 879 880 code = frame->tf_err; 881 type = frame->tf_trapno; 882 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 883 884 if (type <= MAX_TRAP_MSG) 885 msg = trap_msg[type]; 886 else 887 msg = "UNKNOWN"; 888 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 889 frame->tf_eflags & PSL_VM ? "vm86" : 890 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 891 #ifdef SMP 892 /* two separate prints in case of a trap on an unmapped page */ 893 printf("cpuid = %d; ", PCPU_GET(cpuid)); 894 printf("apic id = %02x\n", PCPU_GET(apic_id)); 895 #endif 896 if (type == T_PAGEFLT) { 897 printf("fault virtual address = 0x%x\n", eva); 898 printf("fault code = %s %s, %s\n", 899 code & PGEX_U ? "user" : "supervisor", 900 code & PGEX_W ? "write" : "read", 901 code & PGEX_P ? "protection violation" : "page not present"); 902 } 903 printf("instruction pointer = 0x%x:0x%x\n", 904 frame->tf_cs & 0xffff, frame->tf_eip); 905 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 906 ss = frame->tf_ss & 0xffff; 907 esp = frame->tf_esp; 908 } else { 909 ss = GSEL(GDATA_SEL, SEL_KPL); 910 esp = (int)&frame->tf_esp; 911 } 912 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 913 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 914 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 915 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 916 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 917 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 918 softseg.ssd_gran); 919 printf("processor eflags = "); 920 if (frame->tf_eflags & PSL_T) 921 printf("trace trap, "); 922 if (frame->tf_eflags & PSL_I) 923 printf("interrupt enabled, "); 924 if (frame->tf_eflags & PSL_NT) 925 printf("nested task, "); 926 if (frame->tf_eflags & PSL_RF) 927 printf("resume, "); 928 if (frame->tf_eflags & PSL_VM) 929 printf("vm86, "); 930 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 931 printf("current process = "); 932 if (curproc) { 933 printf("%lu (%s)\n", (u_long)curproc->p_pid, curproc->p_comm); 934 } else { 935 printf("Idle\n"); 936 } 937 938 #ifdef KDB 939 if (debugger_on_panic || kdb_active) { 940 frame->tf_err = eva; /* smuggle fault address to ddb */ 941 if (kdb_trap(type, 0, frame)) { 942 frame->tf_err = code; /* restore error code */ 943 return; 944 } 945 frame->tf_err = code; /* restore error code */ 946 } 947 #endif 948 printf("trap number = %d\n", type); 949 if (type <= MAX_TRAP_MSG) 950 panic("%s", trap_msg[type]); 951 else 952 panic("unknown/reserved trap"); 953 } 954 955 /* 956 * Double fault handler. Called when a fault occurs while writing 957 * a frame for a trap/exception onto the stack. This usually occurs 958 * when the stack overflows (such is the case with infinite recursion, 959 * for example). 960 * 961 * XXX Note that the current PTD gets replaced by IdlePTD when the 962 * task switch occurs. This means that the stack that was active at 963 * the time of the double fault is not available at <kstack> unless 964 * the machine was idle when the double fault occurred. The downside 965 * of this is that "trace <ebp>" in ddb won't work. 966 */ 967 void 968 dblfault_handler() 969 { 970 #ifdef KDTRACE_HOOKS 971 if (dtrace_doubletrap_func != NULL) 972 (*dtrace_doubletrap_func)(); 973 #endif 974 printf("\nFatal double fault:\n"); 975 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip)); 976 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp)); 977 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp)); 978 #ifdef SMP 979 /* two separate prints in case of a trap on an unmapped page */ 980 printf("cpuid = %d; ", PCPU_GET(cpuid)); 981 printf("apic id = %02x\n", PCPU_GET(apic_id)); 982 #endif 983 panic("double fault"); 984 } 985 986 /* 987 * syscall - system call request C handler 988 * 989 * A system call is essentially treated as a trap. 990 */ 991 void 992 syscall(struct trapframe *frame) 993 { 994 caddr_t params; 995 struct sysent *callp; 996 struct thread *td = curthread; 997 struct proc *p = td->td_proc; 998 register_t orig_tf_eflags; 999 int error; 1000 int narg; 1001 int args[8]; 1002 u_int code; 1003 ksiginfo_t ksi; 1004 1005 PCPU_INC(cnt.v_syscall); 1006 1007 #ifdef DIAGNOSTIC 1008 if (ISPL(frame->tf_cs) != SEL_UPL) { 1009 panic("syscall"); 1010 /* NOT REACHED */ 1011 } 1012 #endif 1013 1014 td->td_pticks = 0; 1015 td->td_frame = frame; 1016 if (td->td_ucred != p->p_ucred) 1017 cred_update_thread(td); 1018 #ifdef KSE 1019 if (p->p_flag & P_SA) 1020 thread_user_enter(td); 1021 #endif 1022 params = (caddr_t)frame->tf_esp + sizeof(int); 1023 code = frame->tf_eax; 1024 orig_tf_eflags = frame->tf_eflags; 1025 1026 if (p->p_sysent->sv_prepsyscall) { 1027 /* 1028 * The prep code is MP aware. 1029 */ 1030 (*p->p_sysent->sv_prepsyscall)(frame, args, &code, ¶ms); 1031 } else { 1032 /* 1033 * Need to check if this is a 32 bit or 64 bit syscall. 1034 * fuword is MP aware. 1035 */ 1036 if (code == SYS_syscall) { 1037 /* 1038 * Code is first argument, followed by actual args. 1039 */ 1040 code = fuword(params); 1041 params += sizeof(int); 1042 } else if (code == SYS___syscall) { 1043 /* 1044 * Like syscall, but code is a quad, so as to maintain 1045 * quad alignment for the rest of the arguments. 1046 */ 1047 code = fuword(params); 1048 params += sizeof(quad_t); 1049 } 1050 } 1051 1052 if (p->p_sysent->sv_mask) 1053 code &= p->p_sysent->sv_mask; 1054 1055 if (code >= p->p_sysent->sv_size) 1056 callp = &p->p_sysent->sv_table[0]; 1057 else 1058 callp = &p->p_sysent->sv_table[code]; 1059 1060 narg = callp->sy_narg; 1061 1062 /* 1063 * copyin and the ktrsyscall()/ktrsysret() code is MP-aware 1064 */ 1065 if (params != NULL && narg != 0) 1066 error = copyin(params, (caddr_t)args, 1067 (u_int)(narg * sizeof(int))); 1068 else 1069 error = 0; 1070 1071 #ifdef KTRACE 1072 if (KTRPOINT(td, KTR_SYSCALL)) 1073 ktrsyscall(code, narg, args); 1074 #endif 1075 1076 CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td, 1077 td->td_proc->p_pid, td->td_proc->p_comm, code); 1078 1079 td->td_syscalls++; 1080 1081 if (error == 0) { 1082 td->td_retval[0] = 0; 1083 td->td_retval[1] = frame->tf_edx; 1084 1085 STOPEVENT(p, S_SCE, narg); 1086 1087 PTRACESTOP_SC(p, td, S_PT_SCE); 1088 1089 #ifdef KDTRACE_HOOKS 1090 /* 1091 * If the systrace module has registered it's probe 1092 * callback and if there is a probe active for the 1093 * syscall 'entry', process the probe. 1094 */ 1095 if (systrace_probe_func != NULL && callp->sy_entry != 0) 1096 (*systrace_probe_func)(callp->sy_entry, code, callp, 1097 args); 1098 #endif 1099 1100 AUDIT_SYSCALL_ENTER(code, td); 1101 error = (*callp->sy_call)(td, args); 1102 AUDIT_SYSCALL_EXIT(error, td); 1103 1104 /* Save the latest error return value. */ 1105 td->td_errno = error; 1106 1107 #ifdef KDTRACE_HOOKS 1108 /* 1109 * If the systrace module has registered it's probe 1110 * callback and if there is a probe active for the 1111 * syscall 'return', process the probe. 1112 */ 1113 if (systrace_probe_func != NULL && callp->sy_return != 0) 1114 (*systrace_probe_func)(callp->sy_return, code, callp, 1115 args); 1116 #endif 1117 } 1118 1119 switch (error) { 1120 case 0: 1121 frame->tf_eax = td->td_retval[0]; 1122 frame->tf_edx = td->td_retval[1]; 1123 frame->tf_eflags &= ~PSL_C; 1124 break; 1125 1126 case ERESTART: 1127 /* 1128 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1129 * int 0x80 is 2 bytes. We saved this in tf_err. 1130 */ 1131 frame->tf_eip -= frame->tf_err; 1132 break; 1133 1134 case EJUSTRETURN: 1135 break; 1136 1137 default: 1138 if (p->p_sysent->sv_errsize) { 1139 if (error >= p->p_sysent->sv_errsize) 1140 error = -1; /* XXX */ 1141 else 1142 error = p->p_sysent->sv_errtbl[error]; 1143 } 1144 frame->tf_eax = error; 1145 frame->tf_eflags |= PSL_C; 1146 break; 1147 } 1148 1149 /* 1150 * Traced syscall. 1151 */ 1152 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) { 1153 frame->tf_eflags &= ~PSL_T; 1154 ksiginfo_init_trap(&ksi); 1155 ksi.ksi_signo = SIGTRAP; 1156 ksi.ksi_code = TRAP_TRACE; 1157 ksi.ksi_addr = (void *)frame->tf_eip; 1158 trapsignal(td, &ksi); 1159 } 1160 1161 /* 1162 * Check for misbehavior. 1163 */ 1164 WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning", 1165 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???"); 1166 KASSERT(td->td_critnest == 0, 1167 ("System call %s returning in a critical section", 1168 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???")); 1169 KASSERT(td->td_locks == 0, 1170 ("System call %s returning with %d locks held", 1171 (code >= 0 && code < SYS_MAXSYSCALL) ? syscallnames[code] : "???", 1172 td->td_locks)); 1173 1174 /* 1175 * Handle reschedule and other end-of-syscall issues 1176 */ 1177 userret(td, frame); 1178 1179 CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td, 1180 td->td_proc->p_pid, td->td_proc->p_comm, code); 1181 1182 #ifdef KTRACE 1183 if (KTRPOINT(td, KTR_SYSRET)) 1184 ktrsysret(code, error, td->td_retval[0]); 1185 #endif 1186 1187 /* 1188 * This works because errno is findable through the 1189 * register set. If we ever support an emulation where this 1190 * is not the case, this code will need to be revisited. 1191 */ 1192 STOPEVENT(p, S_SCX, code); 1193 1194 PTRACESTOP_SC(p, td, S_PT_SCX); 1195 } 1196
Cache object: 6a1423934204a2b164d4d2581d263995
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]