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/amd64/amd64/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/10.4/sys/amd64/amd64/trap.c 333371 2018-05-08 17:12:10Z gordon $"); 42 43 /* 44 * AMD64 Trap and System call handling 45 */ 46 47 #include "opt_clock.h" 48 #include "opt_compat.h" 49 #include "opt_cpu.h" 50 #include "opt_hwpmc_hooks.h" 51 #include "opt_isa.h" 52 #include "opt_kdb.h" 53 #include "opt_kdtrace.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 80 #include <vm/vm.h> 81 #include <vm/vm_param.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_kern.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_extern.h> 87 88 #include <machine/cpu.h> 89 #include <machine/intr_machdep.h> 90 #include <x86/mca.h> 91 #include <machine/md_var.h> 92 #include <machine/pcb.h> 93 #ifdef SMP 94 #include <machine/smp.h> 95 #endif 96 #include <machine/tss.h> 97 98 #ifdef KDTRACE_HOOKS 99 #include <sys/dtrace_bsd.h> 100 #endif 101 102 extern inthand_t IDTVEC(bpt), IDTVEC(dbg), IDTVEC(fast_syscall), 103 IDTVEC(fast_syscall32), IDTVEC(int0x80_syscall); 104 105 extern void trap(struct trapframe *frame); 106 extern void syscall(struct trapframe *frame); 107 void dblfault_handler(struct trapframe *frame); 108 109 static int trap_pfault(struct trapframe *, int); 110 static void trap_fatal(struct trapframe *, vm_offset_t); 111 112 #define MAX_TRAP_MSG 32 113 static char *trap_msg[] = { 114 "", /* 0 unused */ 115 "privileged instruction fault", /* 1 T_PRIVINFLT */ 116 "", /* 2 unused */ 117 "breakpoint instruction fault", /* 3 T_BPTFLT */ 118 "", /* 4 unused */ 119 "", /* 5 unused */ 120 "arithmetic trap", /* 6 T_ARITHTRAP */ 121 "", /* 7 unused */ 122 "", /* 8 unused */ 123 "general protection fault", /* 9 T_PROTFLT */ 124 "trace trap", /* 10 T_TRCTRAP */ 125 "", /* 11 unused */ 126 "page fault", /* 12 T_PAGEFLT */ 127 "", /* 13 unused */ 128 "alignment fault", /* 14 T_ALIGNFLT */ 129 "", /* 15 unused */ 130 "", /* 16 unused */ 131 "", /* 17 unused */ 132 "integer divide fault", /* 18 T_DIVIDE */ 133 "non-maskable interrupt trap", /* 19 T_NMI */ 134 "overflow trap", /* 20 T_OFLOW */ 135 "FPU bounds check fault", /* 21 T_BOUND */ 136 "FPU device not available", /* 22 T_DNA */ 137 "double fault", /* 23 T_DOUBLEFLT */ 138 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 139 "invalid TSS fault", /* 25 T_TSSFLT */ 140 "segment not present fault", /* 26 T_SEGNPFLT */ 141 "stack fault", /* 27 T_STKFLT */ 142 "machine check trap", /* 28 T_MCHK */ 143 "SIMD floating-point exception", /* 29 T_XMMFLT */ 144 "reserved (unknown) fault", /* 30 T_RESERVED */ 145 "", /* 31 unused (reserved) */ 146 "DTrace pid return trap", /* 32 T_DTRACE_RET */ 147 }; 148 149 #ifdef KDB 150 static int kdb_on_nmi = 1; 151 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW, 152 &kdb_on_nmi, 0, "Go to KDB on NMI"); 153 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi); 154 #endif 155 static int panic_on_nmi = 1; 156 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 157 &panic_on_nmi, 0, "Panic on NMI"); 158 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi); 159 static int prot_fault_translation; 160 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW, 161 &prot_fault_translation, 0, 162 "Select signal to deliver on protection fault"); 163 static int uprintf_signal; 164 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW, 165 &uprintf_signal, 0, 166 "Print debugging information on trap signal to ctty"); 167 168 /* 169 * Exception, fault, and trap interface to the FreeBSD kernel. 170 * This common code is called from assembly language IDT gate entry 171 * routines that prepare a suitable stack frame, and restore this 172 * frame after the exception has been processed. 173 */ 174 175 void 176 trap(struct trapframe *frame) 177 { 178 #ifdef KDTRACE_HOOKS 179 struct reg regs; 180 #endif 181 struct thread *td = curthread; 182 struct proc *p = td->td_proc; 183 int i = 0, ucode = 0, code; 184 u_int type; 185 register_t addr = 0; 186 ksiginfo_t ksi; 187 188 PCPU_INC(cnt.v_trap); 189 type = frame->tf_trapno; 190 191 #ifdef SMP 192 /* Handler for NMI IPIs used for stopping CPUs. */ 193 if (type == T_NMI) { 194 if (ipi_nmi_handler() == 0) 195 goto out; 196 } 197 #endif /* SMP */ 198 199 #ifdef KDB 200 if (kdb_active) { 201 kdb_reenter(); 202 goto out; 203 } 204 #endif 205 206 if (type == T_RESERVED) { 207 trap_fatal(frame, 0); 208 goto out; 209 } 210 211 #ifdef HWPMC_HOOKS 212 /* 213 * CPU PMCs interrupt using an NMI. If the PMC module is 214 * active, pass the 'rip' value to the PMC module's interrupt 215 * handler. A return value of '1' from the handler means that 216 * the NMI was handled by it and we can return immediately. 217 */ 218 if (type == T_NMI && pmc_intr && 219 (*pmc_intr)(PCPU_GET(cpuid), frame)) 220 goto out; 221 #endif 222 223 if (type == T_MCHK) { 224 mca_intr(); 225 goto out; 226 } 227 228 #ifdef KDTRACE_HOOKS 229 /* 230 * A trap can occur while DTrace executes a probe. Before 231 * executing the probe, DTrace blocks re-scheduling and sets 232 * a flag in its per-cpu flags to indicate that it doesn't 233 * want to fault. On returning from the probe, the no-fault 234 * flag is cleared and finally re-scheduling is enabled. 235 */ 236 if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type)) 237 goto out; 238 #endif 239 240 if ((frame->tf_rflags & PSL_I) == 0) { 241 /* 242 * Buggy application or kernel code has disabled 243 * interrupts and then trapped. Enabling interrupts 244 * now is wrong, but it is better than running with 245 * interrupts disabled until they are accidentally 246 * enabled later. 247 */ 248 if (ISPL(frame->tf_cs) == SEL_UPL) 249 uprintf( 250 "pid %ld (%s): trap %d with interrupts disabled\n", 251 (long)curproc->p_pid, curthread->td_name, type); 252 else if (type != T_NMI && type != T_BPTFLT && 253 type != T_TRCTRAP) { 254 /* 255 * XXX not quite right, since this may be for a 256 * multiple fault in user mode. 257 */ 258 printf("kernel trap %d with interrupts disabled\n", 259 type); 260 261 /* 262 * We shouldn't enable interrupts while holding a 263 * spin lock. 264 */ 265 if (td->td_md.md_spinlock_count == 0) 266 enable_intr(); 267 } 268 } 269 270 code = frame->tf_err; 271 272 if (ISPL(frame->tf_cs) == SEL_UPL) { 273 /* user trap */ 274 275 td->td_pticks = 0; 276 td->td_frame = frame; 277 addr = frame->tf_rip; 278 if (td->td_ucred != p->p_ucred) 279 cred_update_thread(td); 280 281 switch (type) { 282 case T_PRIVINFLT: /* privileged instruction fault */ 283 i = SIGILL; 284 ucode = ILL_PRVOPC; 285 break; 286 287 case T_BPTFLT: /* bpt instruction fault */ 288 case T_TRCTRAP: /* trace trap */ 289 enable_intr(); 290 #ifdef KDTRACE_HOOKS 291 if (type == T_BPTFLT) { 292 fill_frame_regs(frame, ®s); 293 if (dtrace_pid_probe_ptr != NULL && 294 dtrace_pid_probe_ptr(®s) == 0) 295 goto out; 296 } 297 #endif 298 frame->tf_rflags &= ~PSL_T; 299 i = SIGTRAP; 300 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 301 break; 302 303 case T_ARITHTRAP: /* arithmetic trap */ 304 ucode = fputrap_x87(); 305 if (ucode == -1) 306 goto userout; 307 i = SIGFPE; 308 break; 309 310 case T_PROTFLT: /* general protection fault */ 311 i = SIGBUS; 312 ucode = BUS_OBJERR; 313 break; 314 case T_STKFLT: /* stack fault */ 315 case T_SEGNPFLT: /* segment not present fault */ 316 i = SIGBUS; 317 ucode = BUS_ADRERR; 318 break; 319 case T_TSSFLT: /* invalid TSS fault */ 320 i = SIGBUS; 321 ucode = BUS_OBJERR; 322 break; 323 case T_ALIGNFLT: 324 i = SIGBUS; 325 ucode = BUS_ADRALN; 326 break; 327 case T_DOUBLEFLT: /* double fault */ 328 default: 329 i = SIGBUS; 330 ucode = BUS_OBJERR; 331 break; 332 333 case T_PAGEFLT: /* page fault */ 334 /* 335 * Emulator can take care about this trap? 336 */ 337 if (*p->p_sysent->sv_trap != NULL && 338 (*p->p_sysent->sv_trap)(td) == 0) 339 goto userout; 340 341 addr = frame->tf_addr; 342 i = trap_pfault(frame, TRUE); 343 if (i == -1) 344 goto userout; 345 if (i == 0) 346 goto user; 347 348 if (i == SIGSEGV) 349 ucode = SEGV_MAPERR; 350 else { 351 if (prot_fault_translation == 0) { 352 /* 353 * Autodetect. 354 * This check also covers the images 355 * without the ABI-tag ELF note. 356 */ 357 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD 358 && p->p_osrel >= P_OSREL_SIGSEGV) { 359 i = SIGSEGV; 360 ucode = SEGV_ACCERR; 361 } else { 362 i = SIGBUS; 363 ucode = BUS_PAGE_FAULT; 364 } 365 } else if (prot_fault_translation == 1) { 366 /* 367 * Always compat mode. 368 */ 369 i = SIGBUS; 370 ucode = BUS_PAGE_FAULT; 371 } else { 372 /* 373 * Always SIGSEGV mode. 374 */ 375 i = SIGSEGV; 376 ucode = SEGV_ACCERR; 377 } 378 } 379 break; 380 381 case T_DIVIDE: /* integer divide fault */ 382 ucode = FPE_INTDIV; 383 i = SIGFPE; 384 break; 385 386 #ifdef DEV_ISA 387 case T_NMI: 388 /* machine/parity/power fail/"kitchen sink" faults */ 389 if (isa_nmi(code) == 0) { 390 #ifdef KDB 391 /* 392 * NMI can be hooked up to a pushbutton 393 * for debugging. 394 */ 395 if (kdb_on_nmi) { 396 printf ("NMI ... going to debugger\n"); 397 kdb_trap(type, 0, frame); 398 } 399 #endif /* KDB */ 400 goto userout; 401 } else if (panic_on_nmi) 402 panic("NMI indicates hardware failure"); 403 goto out; 404 #endif /* DEV_ISA */ 405 406 case T_OFLOW: /* integer overflow fault */ 407 ucode = FPE_INTOVF; 408 i = SIGFPE; 409 break; 410 411 case T_BOUND: /* bounds check fault */ 412 ucode = FPE_FLTSUB; 413 i = SIGFPE; 414 break; 415 416 case T_DNA: 417 /* transparent fault (due to context switch "late") */ 418 KASSERT(PCB_USER_FPU(td->td_pcb), 419 ("kernel FPU ctx has leaked")); 420 fpudna(); 421 goto userout; 422 423 case T_FPOPFLT: /* FPU operand fetch fault */ 424 ucode = ILL_COPROC; 425 i = SIGILL; 426 break; 427 428 case T_XMMFLT: /* SIMD floating-point exception */ 429 ucode = fputrap_sse(); 430 if (ucode == -1) 431 goto userout; 432 i = SIGFPE; 433 break; 434 #ifdef KDTRACE_HOOKS 435 case T_DTRACE_RET: 436 enable_intr(); 437 fill_frame_regs(frame, ®s); 438 if (dtrace_return_probe_ptr != NULL && 439 dtrace_return_probe_ptr(®s) == 0) 440 goto out; 441 goto userout; 442 #endif 443 } 444 } else { 445 /* kernel trap */ 446 447 KASSERT(cold || td->td_ucred != NULL, 448 ("kernel trap doesn't have ucred")); 449 switch (type) { 450 case T_PAGEFLT: /* page fault */ 451 (void) trap_pfault(frame, FALSE); 452 goto out; 453 454 case T_DNA: 455 if (PCB_USER_FPU(td->td_pcb)) 456 panic("Unregistered use of FPU in kernel"); 457 fpudna(); 458 goto out; 459 460 case T_ARITHTRAP: /* arithmetic trap */ 461 case T_XMMFLT: /* SIMD floating-point exception */ 462 case T_FPOPFLT: /* FPU operand fetch fault */ 463 /* 464 * For now, supporting kernel handler 465 * registration for FPU traps is overkill. 466 */ 467 trap_fatal(frame, 0); 468 goto out; 469 470 case T_STKFLT: /* stack fault */ 471 case T_PROTFLT: /* general protection fault */ 472 case T_SEGNPFLT: /* segment not present fault */ 473 if (td->td_intr_nesting_level != 0) 474 break; 475 476 /* 477 * Invalid segment selectors and out of bounds 478 * %rip's and %rsp's can be set up in user mode. 479 * This causes a fault in kernel mode when the 480 * kernel tries to return to user mode. We want 481 * to get this fault so that we can fix the 482 * problem here and not have to check all the 483 * selectors and pointers when the user changes 484 * them. 485 */ 486 if (frame->tf_rip == (long)doreti_iret) { 487 frame->tf_rip = (long)doreti_iret_fault; 488 goto out; 489 } 490 if (frame->tf_rip == (long)ld_ds) { 491 frame->tf_rip = (long)ds_load_fault; 492 goto out; 493 } 494 if (frame->tf_rip == (long)ld_es) { 495 frame->tf_rip = (long)es_load_fault; 496 goto out; 497 } 498 if (frame->tf_rip == (long)ld_fs) { 499 frame->tf_rip = (long)fs_load_fault; 500 goto out; 501 } 502 if (frame->tf_rip == (long)ld_gs) { 503 frame->tf_rip = (long)gs_load_fault; 504 goto out; 505 } 506 if (frame->tf_rip == (long)ld_gsbase) { 507 frame->tf_rip = (long)gsbase_load_fault; 508 goto out; 509 } 510 if (frame->tf_rip == (long)ld_fsbase) { 511 frame->tf_rip = (long)fsbase_load_fault; 512 goto out; 513 } 514 if (curpcb->pcb_onfault != NULL) { 515 frame->tf_rip = (long)curpcb->pcb_onfault; 516 goto out; 517 } 518 break; 519 520 case T_TSSFLT: 521 /* 522 * PSL_NT can be set in user mode and isn't cleared 523 * automatically when the kernel is entered. This 524 * causes a TSS fault when the kernel attempts to 525 * `iret' because the TSS link is uninitialized. We 526 * want to get this fault so that we can fix the 527 * problem here and not every time the kernel is 528 * entered. 529 */ 530 if (frame->tf_rflags & PSL_NT) { 531 frame->tf_rflags &= ~PSL_NT; 532 goto out; 533 } 534 break; 535 536 case T_TRCTRAP: /* trace trap */ 537 /* 538 * Ignore debug register trace traps due to 539 * accesses in the user's address space, which 540 * can happen under several conditions such as 541 * if a user sets a watchpoint on a buffer and 542 * then passes that buffer to a system call. 543 * We still want to get TRCTRAPS for addresses 544 * in kernel space because that is useful when 545 * debugging the kernel. 546 */ 547 if (user_dbreg_trap()) { 548 /* 549 * Reset breakpoint bits because the 550 * processor doesn't 551 */ 552 /* XXX check upper bits here */ 553 load_dr6(rdr6() & 0xfffffff0); 554 goto out; 555 } 556 557 /* 558 * Malicious user code can configure a debug 559 * register watchpoint to trap on data access 560 * to the top of stack and then execute 'pop 561 * %ss; int 3'. Due to exception deferral for 562 * 'pop %ss', the CPU will not interrupt 'int 563 * 3' to raise the DB# exception for the debug 564 * register but will postpone the DB# until 565 * execution of the first instruction of the 566 * BP# handler (in kernel mode). Normally the 567 * previous check would ignore DB# exceptions 568 * for watchpoints on user addresses raised in 569 * kernel mode. However, some CPU errata 570 * include cases where DB# exceptions do not 571 * properly set bits in %dr6, e.g. Haswell 572 * HSD23 and Skylake-X SKZ24. 573 * 574 * A deferred DB# can also be raised on the 575 * first instructions of system call entry 576 * points or single-step traps via similar use 577 * of 'pop %ss' or 'mov xxx, %ss'. 578 */ 579 if (frame->tf_rip == (uintptr_t)IDTVEC(fast_syscall) || 580 #ifdef COMPAT_FREEBSD32 581 frame->tf_rip == 582 (uintptr_t)IDTVEC(int0x80_syscall) || 583 #endif 584 frame->tf_rip == (uintptr_t)IDTVEC(bpt) || 585 frame->tf_rip == (uintptr_t)IDTVEC(dbg) || 586 /* Needed for AMD. */ 587 frame->tf_rip == (uintptr_t)IDTVEC(fast_syscall32)) 588 return; 589 /* 590 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 591 */ 592 case T_BPTFLT: 593 /* 594 * If KDB is enabled, let it handle the debugger trap. 595 * Otherwise, debugger traps "can't happen". 596 */ 597 #ifdef KDB 598 if (kdb_trap(type, 0, frame)) 599 goto out; 600 #endif 601 break; 602 603 #ifdef DEV_ISA 604 case T_NMI: 605 /* machine/parity/power fail/"kitchen sink" faults */ 606 if (isa_nmi(code) == 0) { 607 #ifdef KDB 608 /* 609 * NMI can be hooked up to a pushbutton 610 * for debugging. 611 */ 612 if (kdb_on_nmi) { 613 printf ("NMI ... going to debugger\n"); 614 kdb_trap(type, 0, frame); 615 } 616 #endif /* KDB */ 617 goto out; 618 } else if (panic_on_nmi == 0) 619 goto out; 620 /* FALLTHROUGH */ 621 #endif /* DEV_ISA */ 622 } 623 624 trap_fatal(frame, 0); 625 goto out; 626 } 627 628 /* Translate fault for emulators (e.g. Linux) */ 629 if (*p->p_sysent->sv_transtrap) 630 i = (*p->p_sysent->sv_transtrap)(i, type); 631 632 ksiginfo_init_trap(&ksi); 633 ksi.ksi_signo = i; 634 ksi.ksi_code = ucode; 635 ksi.ksi_trapno = type; 636 ksi.ksi_addr = (void *)addr; 637 if (uprintf_signal) { 638 uprintf("pid %d comm %s: signal %d err %lx code %d type %d " 639 "addr 0x%lx rsp 0x%lx rip 0x%lx " 640 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n", 641 p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr, 642 frame->tf_rsp, frame->tf_rip, 643 fubyte((void *)(frame->tf_rip + 0)), 644 fubyte((void *)(frame->tf_rip + 1)), 645 fubyte((void *)(frame->tf_rip + 2)), 646 fubyte((void *)(frame->tf_rip + 3)), 647 fubyte((void *)(frame->tf_rip + 4)), 648 fubyte((void *)(frame->tf_rip + 5)), 649 fubyte((void *)(frame->tf_rip + 6)), 650 fubyte((void *)(frame->tf_rip + 7))); 651 } 652 KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled")); 653 trapsignal(td, &ksi); 654 655 user: 656 userret(td, frame); 657 KASSERT(PCB_USER_FPU(td->td_pcb), 658 ("Return from trap with kernel FPU ctx leaked")); 659 userout: 660 out: 661 return; 662 } 663 664 static int 665 trap_pfault(frame, usermode) 666 struct trapframe *frame; 667 int usermode; 668 { 669 vm_offset_t va; 670 struct vmspace *vm; 671 vm_map_t map; 672 int rv = 0; 673 vm_prot_t ftype; 674 struct thread *td = curthread; 675 struct proc *p = td->td_proc; 676 vm_offset_t eva = frame->tf_addr; 677 678 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) { 679 /* 680 * Due to both processor errata and lazy TLB invalidation when 681 * access restrictions are removed from virtual pages, memory 682 * accesses that are allowed by the physical mapping layer may 683 * nonetheless cause one spurious page fault per virtual page. 684 * When the thread is executing a "no faulting" section that 685 * is bracketed by vm_fault_{disable,enable}_pagefaults(), 686 * every page fault is treated as a spurious page fault, 687 * unless it accesses the same virtual address as the most 688 * recent page fault within the same "no faulting" section. 689 */ 690 if (td->td_md.md_spurflt_addr != eva || 691 (td->td_pflags & TDP_RESETSPUR) != 0) { 692 /* 693 * Do nothing to the TLB. A stale TLB entry is 694 * flushed automatically by a page fault. 695 */ 696 td->td_md.md_spurflt_addr = eva; 697 td->td_pflags &= ~TDP_RESETSPUR; 698 return (0); 699 } 700 } else { 701 /* 702 * If we get a page fault while in a critical section, then 703 * it is most likely a fatal kernel page fault. The kernel 704 * is already going to panic trying to get a sleep lock to 705 * do the VM lookup, so just consider it a fatal trap so the 706 * kernel can print out a useful trap message and even get 707 * to the debugger. 708 * 709 * If we get a page fault while holding a non-sleepable 710 * lock, then it is most likely a fatal kernel page fault. 711 * If WITNESS is enabled, then it's going to whine about 712 * bogus LORs with various VM locks, so just skip to the 713 * fatal trap handling directly. 714 */ 715 if (td->td_critnest != 0 || 716 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL, 717 "Kernel page fault") != 0) { 718 trap_fatal(frame, eva); 719 return (-1); 720 } 721 } 722 va = trunc_page(eva); 723 if (va >= VM_MIN_KERNEL_ADDRESS) { 724 /* 725 * Don't allow user-mode faults in kernel address space. 726 */ 727 if (usermode) 728 goto nogo; 729 730 map = kernel_map; 731 } else { 732 /* 733 * This is a fault on non-kernel virtual memory. If either 734 * p or p->p_vmspace is NULL, then the fault is fatal. 735 */ 736 if (p == NULL || (vm = p->p_vmspace) == NULL) 737 goto nogo; 738 739 map = &vm->vm_map; 740 741 /* 742 * When accessing a usermode address, kernel must be 743 * ready to accept the page fault, and provide a 744 * handling routine. Since accessing the address 745 * without the handler is a bug, do not try to handle 746 * it normally, and panic immediately. 747 */ 748 if (!usermode && (td->td_intr_nesting_level != 0 || 749 curpcb->pcb_onfault == NULL)) { 750 trap_fatal(frame, eva); 751 return (-1); 752 } 753 } 754 755 /* 756 * If the trap was caused by errant bits in the PTE then panic. 757 */ 758 if (frame->tf_err & PGEX_RSV) { 759 trap_fatal(frame, eva); 760 return (-1); 761 } 762 763 /* 764 * PGEX_I is defined only if the execute disable bit capability is 765 * supported and enabled. 766 */ 767 if (frame->tf_err & PGEX_W) 768 ftype = VM_PROT_WRITE; 769 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 770 ftype = VM_PROT_EXECUTE; 771 else 772 ftype = VM_PROT_READ; 773 774 if (map != kernel_map) { 775 /* 776 * Keep swapout from messing with us during this 777 * critical time. 778 */ 779 PROC_LOCK(p); 780 ++p->p_lock; 781 PROC_UNLOCK(p); 782 783 /* Fault in the user page: */ 784 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 785 786 PROC_LOCK(p); 787 --p->p_lock; 788 PROC_UNLOCK(p); 789 } else { 790 /* 791 * Don't have to worry about process locking or stacks in the 792 * kernel. 793 */ 794 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 795 } 796 if (rv == KERN_SUCCESS) { 797 #ifdef HWPMC_HOOKS 798 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 799 PMC_SOFT_CALL_TF( , , page_fault, all, frame); 800 if (ftype == VM_PROT_READ) 801 PMC_SOFT_CALL_TF( , , page_fault, read, 802 frame); 803 else 804 PMC_SOFT_CALL_TF( , , page_fault, write, 805 frame); 806 } 807 #endif 808 return (0); 809 } 810 nogo: 811 if (!usermode) { 812 if (td->td_intr_nesting_level == 0 && 813 curpcb->pcb_onfault != NULL) { 814 frame->tf_rip = (long)curpcb->pcb_onfault; 815 return (0); 816 } 817 trap_fatal(frame, eva); 818 return (-1); 819 } 820 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 821 } 822 823 static void 824 trap_fatal(frame, eva) 825 struct trapframe *frame; 826 vm_offset_t eva; 827 { 828 int code, ss; 829 u_int type; 830 long esp; 831 struct soft_segment_descriptor softseg; 832 char *msg; 833 834 code = frame->tf_err; 835 type = frame->tf_trapno; 836 sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)], 837 &softseg); 838 839 if (type <= MAX_TRAP_MSG) 840 msg = trap_msg[type]; 841 else 842 msg = "UNKNOWN"; 843 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 844 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 845 #ifdef SMP 846 /* two separate prints in case of a trap on an unmapped page */ 847 printf("cpuid = %d; ", PCPU_GET(cpuid)); 848 printf("apic id = %02x\n", PCPU_GET(apic_id)); 849 #endif 850 if (type == T_PAGEFLT) { 851 printf("fault virtual address = 0x%lx\n", eva); 852 printf("fault code = %s %s %s, %s\n", 853 code & PGEX_U ? "user" : "supervisor", 854 code & PGEX_W ? "write" : "read", 855 code & PGEX_I ? "instruction" : "data", 856 code & PGEX_RSV ? "reserved bits in PTE" : 857 code & PGEX_P ? "protection violation" : "page not present"); 858 } 859 printf("instruction pointer = 0x%lx:0x%lx\n", 860 frame->tf_cs & 0xffff, frame->tf_rip); 861 if (ISPL(frame->tf_cs) == SEL_UPL) { 862 ss = frame->tf_ss & 0xffff; 863 esp = frame->tf_rsp; 864 } else { 865 ss = GSEL(GDATA_SEL, SEL_KPL); 866 esp = (long)&frame->tf_rsp; 867 } 868 printf("stack pointer = 0x%x:0x%lx\n", ss, esp); 869 printf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 870 printf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 871 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 872 printf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 873 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 874 softseg.ssd_gran); 875 printf("processor eflags = "); 876 if (frame->tf_rflags & PSL_T) 877 printf("trace trap, "); 878 if (frame->tf_rflags & PSL_I) 879 printf("interrupt enabled, "); 880 if (frame->tf_rflags & PSL_NT) 881 printf("nested task, "); 882 if (frame->tf_rflags & PSL_RF) 883 printf("resume, "); 884 printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 885 printf("current process = %d (%s)\n", 886 curproc->p_pid, curthread->td_name); 887 888 #ifdef KDB 889 if (debugger_on_panic || kdb_active) 890 if (kdb_trap(type, 0, frame)) 891 return; 892 #endif 893 printf("trap number = %d\n", type); 894 if (type <= MAX_TRAP_MSG) 895 panic("%s", trap_msg[type]); 896 else 897 panic("unknown/reserved trap"); 898 } 899 900 /* 901 * Double fault handler. Called when a fault occurs while writing 902 * a frame for a trap/exception onto the stack. This usually occurs 903 * when the stack overflows (such is the case with infinite recursion, 904 * for example). 905 */ 906 void 907 dblfault_handler(struct trapframe *frame) 908 { 909 #ifdef KDTRACE_HOOKS 910 if (dtrace_doubletrap_func != NULL) 911 (*dtrace_doubletrap_func)(); 912 #endif 913 printf("\nFatal double fault\n"); 914 printf("rip = 0x%lx\n", frame->tf_rip); 915 printf("rsp = 0x%lx\n", frame->tf_rsp); 916 printf("rbp = 0x%lx\n", frame->tf_rbp); 917 #ifdef SMP 918 /* two separate prints in case of a trap on an unmapped page */ 919 printf("cpuid = %d; ", PCPU_GET(cpuid)); 920 printf("apic id = %02x\n", PCPU_GET(apic_id)); 921 #endif 922 panic("double fault"); 923 } 924 925 int 926 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa) 927 { 928 struct proc *p; 929 struct trapframe *frame; 930 register_t *argp; 931 caddr_t params; 932 int reg, regcnt, error; 933 934 p = td->td_proc; 935 frame = td->td_frame; 936 reg = 0; 937 regcnt = 6; 938 939 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 940 sa->code = frame->tf_rax; 941 942 if (sa->code == SYS_syscall || sa->code == SYS___syscall) { 943 sa->code = frame->tf_rdi; 944 reg++; 945 regcnt--; 946 } 947 if (p->p_sysent->sv_mask) 948 sa->code &= p->p_sysent->sv_mask; 949 950 if (sa->code >= p->p_sysent->sv_size) 951 sa->callp = &p->p_sysent->sv_table[0]; 952 else 953 sa->callp = &p->p_sysent->sv_table[sa->code]; 954 955 sa->narg = sa->callp->sy_narg; 956 KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]), 957 ("Too many syscall arguments!")); 958 error = 0; 959 argp = &frame->tf_rdi; 960 argp += reg; 961 bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt); 962 if (sa->narg > regcnt) { 963 KASSERT(params != NULL, ("copyin args with no params!")); 964 error = copyin(params, &sa->args[regcnt], 965 (sa->narg - regcnt) * sizeof(sa->args[0])); 966 } 967 968 if (error == 0) { 969 td->td_retval[0] = 0; 970 td->td_retval[1] = frame->tf_rdx; 971 } 972 973 return (error); 974 } 975 976 #include "../../kern/subr_syscall.c" 977 978 /* 979 * System call handler for native binaries. The trap frame is already 980 * set up by the assembler trampoline and a pointer to it is saved in 981 * td_frame. 982 */ 983 void 984 amd64_syscall(struct thread *td, int traced) 985 { 986 struct syscall_args sa; 987 int error; 988 ksiginfo_t ksi; 989 990 #ifdef DIAGNOSTIC 991 if (ISPL(td->td_frame->tf_cs) != SEL_UPL) { 992 panic("syscall"); 993 /* NOT REACHED */ 994 } 995 #endif 996 error = syscallenter(td, &sa); 997 998 /* 999 * Traced syscall. 1000 */ 1001 if (__predict_false(traced)) { 1002 td->td_frame->tf_rflags &= ~PSL_T; 1003 ksiginfo_init_trap(&ksi); 1004 ksi.ksi_signo = SIGTRAP; 1005 ksi.ksi_code = TRAP_TRACE; 1006 ksi.ksi_addr = (void *)td->td_frame->tf_rip; 1007 trapsignal(td, &ksi); 1008 } 1009 1010 KASSERT(PCB_USER_FPU(td->td_pcb), 1011 ("System call %s returning with kernel FPU ctx leaked", 1012 syscallname(td->td_proc, sa.code))); 1013 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td), 1014 ("System call %s returning with mangled pcb_save", 1015 syscallname(td->td_proc, sa.code))); 1016 1017 syscallret(td, error, &sa); 1018 1019 /* 1020 * If the user-supplied value of %rip is not a canonical 1021 * address, then some CPUs will trigger a ring 0 #GP during 1022 * the sysret instruction. However, the fault handler would 1023 * execute in ring 0 with the user's %gs and %rsp which would 1024 * not be safe. Instead, use the full return path which 1025 * catches the problem safely. 1026 */ 1027 if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS) 1028 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 1029 }
Cache object: 8167abb34c2cc86935f223ff72b59dc1
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]