1 /*- 2 * SPDX-License-Identifier: BSD-4-Clause 3 * 4 * Copyright (C) 1994, David Greenman 5 * Copyright (c) 1990, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the University of Utah, and William Jolitz. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 40 */ 41 42 #include <sys/cdefs.h> 43 __FBSDID("$FreeBSD$"); 44 45 /* 46 * 386 Trap and System call handling 47 */ 48 49 #include "opt_clock.h" 50 #include "opt_compat.h" 51 #include "opt_cpu.h" 52 #include "opt_hwpmc_hooks.h" 53 #include "opt_isa.h" 54 #include "opt_kdb.h" 55 #include "opt_stack.h" 56 #include "opt_trap.h" 57 58 #include <sys/param.h> 59 #include <sys/bus.h> 60 #include <sys/systm.h> 61 #include <sys/proc.h> 62 #include <sys/pioctl.h> 63 #include <sys/ptrace.h> 64 #include <sys/kdb.h> 65 #include <sys/kernel.h> 66 #include <sys/ktr.h> 67 #include <sys/lock.h> 68 #include <sys/mutex.h> 69 #include <sys/resourcevar.h> 70 #include <sys/signalvar.h> 71 #include <sys/syscall.h> 72 #include <sys/sysctl.h> 73 #include <sys/sysent.h> 74 #include <sys/uio.h> 75 #include <sys/vmmeter.h> 76 #ifdef HWPMC_HOOKS 77 #include <sys/pmckern.h> 78 PMC_SOFT_DEFINE( , , page_fault, all); 79 PMC_SOFT_DEFINE( , , page_fault, read); 80 PMC_SOFT_DEFINE( , , page_fault, write); 81 #endif 82 #include <security/audit/audit.h> 83 84 #include <vm/vm.h> 85 #include <vm/vm_param.h> 86 #include <vm/pmap.h> 87 #include <vm/vm_kern.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_page.h> 90 #include <vm/vm_extern.h> 91 92 #include <machine/cpu.h> 93 #include <machine/intr_machdep.h> 94 #include <x86/mca.h> 95 #include <machine/md_var.h> 96 #include <machine/pcb.h> 97 #ifdef SMP 98 #include <machine/smp.h> 99 #endif 100 #include <machine/stack.h> 101 #include <machine/trap.h> 102 #include <machine/tss.h> 103 #include <machine/vm86.h> 104 105 #ifdef POWERFAIL_NMI 106 #include <sys/syslog.h> 107 #include <machine/clock.h> 108 #endif 109 110 #ifdef KDTRACE_HOOKS 111 #include <sys/dtrace_bsd.h> 112 #endif 113 114 void trap(struct trapframe *frame); 115 void syscall(struct trapframe *frame); 116 117 static int trap_pfault(struct trapframe *, int, vm_offset_t); 118 static void trap_fatal(struct trapframe *, vm_offset_t); 119 void dblfault_handler(void); 120 121 extern inthand_t IDTVEC(bpt), IDTVEC(dbg), IDTVEC(int0x80_syscall); 122 123 #define MAX_TRAP_MSG 32 124 125 struct trap_data { 126 bool ei; 127 const char *msg; 128 }; 129 130 static const struct trap_data trap_data[] = { 131 [T_PRIVINFLT] = { .ei = true, .msg = "privileged instruction fault" }, 132 [T_BPTFLT] = { .ei = false, .msg = "breakpoint instruction fault" }, 133 [T_ARITHTRAP] = { .ei = true, .msg = "arithmetic trap" }, 134 [T_PROTFLT] = { .ei = true, .msg = "general protection fault" }, 135 [T_TRCTRAP] = { .ei = false, .msg = "debug exception" }, 136 [T_PAGEFLT] = { .ei = true, .msg = "page fault" }, 137 [T_ALIGNFLT] = { .ei = true, .msg = "alignment fault" }, 138 [T_DIVIDE] = { .ei = true, .msg = "integer divide fault" }, 139 [T_NMI] = { .ei = false, .msg = "non-maskable interrupt trap" }, 140 [T_OFLOW] = { .ei = true, .msg = "overflow trap" }, 141 [T_BOUND] = { .ei = true, .msg = "FPU bounds check fault" }, 142 [T_DNA] = { .ei = true, .msg = "FPU device not available" }, 143 [T_DOUBLEFLT] = { .ei = false, .msg = "double fault" }, 144 [T_FPOPFLT] = { .ei = true, .msg = "FPU operand fetch fault" }, 145 [T_TSSFLT] = { .ei = true, .msg = "invalid TSS fault" }, 146 [T_SEGNPFLT] = { .ei = true, .msg = "segment not present fault" }, 147 [T_STKFLT] = { .ei = true, .msg = "stack fault" }, 148 [T_MCHK] = { .ei = true, .msg = "machine check trap" }, 149 [T_XMMFLT] = { .ei = true, .msg = "SIMD floating-point exception" }, 150 [T_DTRACE_RET] ={ .ei = true, .msg = "DTrace pid return trap" }, 151 }; 152 153 static bool 154 trap_enable_intr(int trapno) 155 { 156 157 MPASS(trapno > 0); 158 if (trapno < nitems(trap_data) && trap_data[trapno].msg != NULL) 159 return (trap_data[trapno].ei); 160 return (false); 161 } 162 163 static const char * 164 trap_msg(int trapno) 165 { 166 const char *res; 167 static const char unkn[] = "UNKNOWN"; 168 169 res = NULL; 170 if (trapno < nitems(trap_data)) 171 res = trap_data[trapno].msg; 172 if (res == NULL) 173 res = unkn; 174 return (res); 175 } 176 177 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 178 int has_f00f_bug = 0; /* Initialized so that it can be patched. */ 179 #endif 180 181 static int prot_fault_translation = 0; 182 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW, 183 &prot_fault_translation, 0, "Select signal to deliver on protection fault"); 184 static int uprintf_signal; 185 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW, 186 &uprintf_signal, 0, 187 "Print debugging information on trap signal to ctty"); 188 189 /* 190 * Exception, fault, and trap interface to the FreeBSD kernel. 191 * This common code is called from assembly language IDT gate entry 192 * routines that prepare a suitable stack frame, and restore this 193 * frame after the exception has been processed. 194 */ 195 196 void 197 trap(struct trapframe *frame) 198 { 199 ksiginfo_t ksi; 200 struct thread *td; 201 struct proc *p; 202 int signo, ucode; 203 u_int type; 204 register_t addr, dr6; 205 vm_offset_t eva; 206 #ifdef POWERFAIL_NMI 207 static int lastalert = 0; 208 #endif 209 210 td = curthread; 211 p = td->td_proc; 212 signo = 0; 213 ucode = 0; 214 addr = 0; 215 dr6 = 0; 216 217 VM_CNT_INC(v_trap); 218 type = frame->tf_trapno; 219 220 KASSERT((read_eflags() & PSL_I) == 0, 221 ("trap: interrupts enabled, type %d frame %p", type, frame)); 222 223 #ifdef SMP 224 /* Handler for NMI IPIs used for stopping CPUs. */ 225 if (type == T_NMI && ipi_nmi_handler() == 0) 226 return; 227 #endif /* SMP */ 228 229 #ifdef KDB 230 if (kdb_active) { 231 kdb_reenter(); 232 return; 233 } 234 #endif 235 236 if (type == T_RESERVED) { 237 trap_fatal(frame, 0); 238 return; 239 } 240 241 if (type == T_NMI) { 242 #ifdef HWPMC_HOOKS 243 /* 244 * CPU PMCs interrupt using an NMI so we check for that first. 245 * If the HWPMC module is active, 'pmc_hook' will point to 246 * the function to be called. A non-zero return value from the 247 * hook means that the NMI was consumed by it and that we can 248 * return immediately. 249 */ 250 if (pmc_intr != NULL && 251 (*pmc_intr)(frame) != 0) 252 return; 253 #endif 254 255 #ifdef STACK 256 if (stack_nmi_handler(frame) != 0) 257 return; 258 #endif 259 } 260 261 if (type == T_MCHK) { 262 mca_intr(); 263 return; 264 } 265 266 #ifdef KDTRACE_HOOKS 267 /* 268 * A trap can occur while DTrace executes a probe. Before 269 * executing the probe, DTrace blocks re-scheduling and sets 270 * a flag in its per-cpu flags to indicate that it doesn't 271 * want to fault. On returning from the probe, the no-fault 272 * flag is cleared and finally re-scheduling is enabled. 273 */ 274 if ((type == T_PROTFLT || type == T_PAGEFLT) && 275 dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type)) 276 return; 277 #endif 278 279 /* 280 * We must not allow context switches until %cr2 is read. 281 * Also, for some Cyrix CPUs, %cr2 is clobbered by interrupts. 282 * All faults use interrupt gates, so %cr2 can be safely read 283 * now, before optional enable of the interrupts below. 284 */ 285 if (type == T_PAGEFLT) 286 eva = rcr2(); 287 288 /* 289 * Buggy application or kernel code has disabled interrupts 290 * and then trapped. Enabling interrupts now is wrong, but it 291 * is better than running with interrupts disabled until they 292 * are accidentally enabled later. 293 */ 294 if ((frame->tf_eflags & PSL_I) == 0 && TRAPF_USERMODE(frame) && 295 (curpcb->pcb_flags & PCB_VM86CALL) == 0) 296 uprintf("pid %ld (%s): trap %d with interrupts disabled\n", 297 (long)curproc->p_pid, curthread->td_name, type); 298 299 /* 300 * Conditionally reenable interrupts. If we hold a spin lock, 301 * then we must not reenable interrupts. This might be a 302 * spurious page fault. 303 */ 304 if (trap_enable_intr(type) && td->td_md.md_spinlock_count == 0 && 305 frame->tf_eip != (int)cpu_switch_load_gs) 306 enable_intr(); 307 308 if (TRAPF_USERMODE(frame) && (curpcb->pcb_flags & PCB_VM86CALL) == 0) { 309 /* user trap */ 310 311 td->td_pticks = 0; 312 td->td_frame = frame; 313 addr = frame->tf_eip; 314 if (td->td_cowgen != p->p_cowgen) 315 thread_cow_update(td); 316 317 switch (type) { 318 case T_PRIVINFLT: /* privileged instruction fault */ 319 signo = SIGILL; 320 ucode = ILL_PRVOPC; 321 break; 322 323 case T_BPTFLT: /* bpt instruction fault */ 324 enable_intr(); 325 #ifdef KDTRACE_HOOKS 326 if (dtrace_pid_probe_ptr != NULL && 327 dtrace_pid_probe_ptr(frame) == 0) 328 return; 329 #endif 330 signo = SIGTRAP; 331 ucode = TRAP_BRKPT; 332 break; 333 334 case T_TRCTRAP: /* debug exception */ 335 enable_intr(); 336 user_trctrap_out: 337 signo = SIGTRAP; 338 ucode = TRAP_TRACE; 339 dr6 = rdr6(); 340 if ((dr6 & DBREG_DR6_BS) != 0) { 341 PROC_LOCK(td->td_proc); 342 if ((td->td_dbgflags & TDB_STEP) != 0) { 343 td->td_frame->tf_eflags &= ~PSL_T; 344 td->td_dbgflags &= ~TDB_STEP; 345 } 346 PROC_UNLOCK(td->td_proc); 347 } 348 break; 349 350 case T_ARITHTRAP: /* arithmetic trap */ 351 ucode = npxtrap_x87(); 352 if (ucode == -1) 353 return; 354 signo = SIGFPE; 355 break; 356 357 /* 358 * The following two traps can happen in vm86 mode, 359 * and, if so, we want to handle them specially. 360 */ 361 case T_PROTFLT: /* general protection fault */ 362 case T_STKFLT: /* stack fault */ 363 if (frame->tf_eflags & PSL_VM) { 364 signo = vm86_emulate((struct vm86frame *)frame); 365 if (signo == SIGTRAP) { 366 load_dr6(rdr6() | 0x4000); 367 goto user_trctrap_out; 368 } 369 if (signo == 0) 370 goto user; 371 break; 372 } 373 signo = SIGBUS; 374 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR; 375 break; 376 case T_SEGNPFLT: /* segment not present fault */ 377 signo = SIGBUS; 378 ucode = BUS_ADRERR; 379 break; 380 case T_TSSFLT: /* invalid TSS fault */ 381 signo = SIGBUS; 382 ucode = BUS_OBJERR; 383 break; 384 case T_ALIGNFLT: 385 signo = SIGBUS; 386 ucode = BUS_ADRALN; 387 break; 388 case T_DOUBLEFLT: /* double fault */ 389 default: 390 signo = SIGBUS; 391 ucode = BUS_OBJERR; 392 break; 393 394 case T_PAGEFLT: /* page fault */ 395 signo = trap_pfault(frame, TRUE, eva); 396 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 397 if (signo == -2) { 398 /* 399 * The f00f hack workaround has triggered, so 400 * treat the fault as an illegal instruction 401 * (T_PRIVINFLT) instead of a page fault. 402 */ 403 type = frame->tf_trapno = T_PRIVINFLT; 404 405 /* Proceed as in that case. */ 406 ucode = ILL_PRVOPC; 407 signo = SIGILL; 408 break; 409 } 410 #endif 411 if (signo == -1) 412 return; 413 if (signo == 0) 414 goto user; 415 416 if (signo == SIGSEGV) 417 ucode = SEGV_MAPERR; 418 else if (prot_fault_translation == 0) { 419 /* 420 * Autodetect. This check also covers 421 * the images without the ABI-tag ELF 422 * note. 423 */ 424 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD && 425 p->p_osrel >= P_OSREL_SIGSEGV) { 426 signo = SIGSEGV; 427 ucode = SEGV_ACCERR; 428 } else { 429 signo = SIGBUS; 430 ucode = T_PAGEFLT; 431 } 432 } else if (prot_fault_translation == 1) { 433 /* 434 * Always compat mode. 435 */ 436 signo = SIGBUS; 437 ucode = T_PAGEFLT; 438 } else { 439 /* 440 * Always SIGSEGV mode. 441 */ 442 signo = SIGSEGV; 443 ucode = SEGV_ACCERR; 444 } 445 addr = eva; 446 break; 447 448 case T_DIVIDE: /* integer divide fault */ 449 ucode = FPE_INTDIV; 450 signo = SIGFPE; 451 break; 452 453 #ifdef DEV_ISA 454 case T_NMI: 455 #ifdef POWERFAIL_NMI 456 #ifndef TIMER_FREQ 457 # define TIMER_FREQ 1193182 458 #endif 459 if (time_second - lastalert > 10) { 460 log(LOG_WARNING, "NMI: power fail\n"); 461 sysbeep(880, hz); 462 lastalert = time_second; 463 } 464 return; 465 #else /* !POWERFAIL_NMI */ 466 nmi_handle_intr(type, frame); 467 return; 468 #endif /* POWERFAIL_NMI */ 469 #endif /* DEV_ISA */ 470 471 case T_OFLOW: /* integer overflow fault */ 472 ucode = FPE_INTOVF; 473 signo = SIGFPE; 474 break; 475 476 case T_BOUND: /* bounds check fault */ 477 ucode = FPE_FLTSUB; 478 signo = SIGFPE; 479 break; 480 481 case T_DNA: 482 KASSERT(PCB_USER_FPU(td->td_pcb), 483 ("kernel FPU ctx has leaked")); 484 /* transparent fault (due to context switch "late") */ 485 if (npxdna()) 486 return; 487 uprintf("pid %d killed due to lack of floating point\n", 488 p->p_pid); 489 signo = SIGKILL; 490 ucode = 0; 491 break; 492 493 case T_FPOPFLT: /* FPU operand fetch fault */ 494 ucode = ILL_COPROC; 495 signo = SIGILL; 496 break; 497 498 case T_XMMFLT: /* SIMD floating-point exception */ 499 ucode = npxtrap_sse(); 500 if (ucode == -1) 501 return; 502 signo = SIGFPE; 503 break; 504 #ifdef KDTRACE_HOOKS 505 case T_DTRACE_RET: 506 enable_intr(); 507 if (dtrace_return_probe_ptr != NULL) 508 dtrace_return_probe_ptr(frame); 509 return; 510 #endif 511 } 512 } else { 513 /* kernel trap */ 514 515 KASSERT(cold || td->td_ucred != NULL, 516 ("kernel trap doesn't have ucred")); 517 switch (type) { 518 case T_PAGEFLT: /* page fault */ 519 (void) trap_pfault(frame, FALSE, eva); 520 return; 521 522 case T_DNA: 523 if (PCB_USER_FPU(td->td_pcb)) 524 panic("Unregistered use of FPU in kernel"); 525 if (npxdna()) 526 return; 527 break; 528 529 case T_ARITHTRAP: /* arithmetic trap */ 530 case T_XMMFLT: /* SIMD floating-point exception */ 531 case T_FPOPFLT: /* FPU operand fetch fault */ 532 /* 533 * XXXKIB for now disable any FPU traps in kernel 534 * handler registration seems to be overkill 535 */ 536 trap_fatal(frame, 0); 537 return; 538 539 /* 540 * The following two traps can happen in 541 * vm86 mode, and, if so, we want to handle 542 * them specially. 543 */ 544 case T_PROTFLT: /* general protection fault */ 545 case T_STKFLT: /* stack fault */ 546 if (frame->tf_eflags & PSL_VM) { 547 signo = vm86_emulate((struct vm86frame *)frame); 548 if (signo == SIGTRAP) { 549 type = T_TRCTRAP; 550 load_dr6(rdr6() | 0x4000); 551 goto kernel_trctrap; 552 } 553 if (signo != 0) 554 /* 555 * returns to original process 556 */ 557 vm86_trap((struct vm86frame *)frame); 558 return; 559 } 560 /* FALL THROUGH */ 561 case T_SEGNPFLT: /* segment not present fault */ 562 if (curpcb->pcb_flags & PCB_VM86CALL) 563 break; 564 565 /* 566 * Invalid %fs's and %gs's can be created using 567 * procfs or PT_SETREGS or by invalidating the 568 * underlying LDT entry. This causes a fault 569 * in kernel mode when the kernel attempts to 570 * switch contexts. Lose the bad context 571 * (XXX) so that we can continue, and generate 572 * a signal. 573 */ 574 if (frame->tf_eip == (int)cpu_switch_load_gs) { 575 curpcb->pcb_gs = 0; 576 #if 0 577 PROC_LOCK(p); 578 kern_psignal(p, SIGBUS); 579 PROC_UNLOCK(p); 580 #endif 581 return; 582 } 583 584 if (td->td_intr_nesting_level != 0) 585 break; 586 587 /* 588 * Invalid segment selectors and out of bounds 589 * %eip's and %esp's can be set up in user mode. 590 * This causes a fault in kernel mode when the 591 * kernel tries to return to user mode. We want 592 * to get this fault so that we can fix the 593 * problem here and not have to check all the 594 * selectors and pointers when the user changes 595 * them. 596 * 597 * N.B. Comparing to long mode, 32-bit mode 598 * does not push %esp on the trap frame, 599 * because iretl faulted while in ring 0. As 600 * the consequence, there is no need to fixup 601 * the stack pointer for doreti_iret_fault, 602 * the fixup and the complimentary trap() call 603 * are executed on the main thread stack, not 604 * on the trampoline stack. 605 */ 606 if (frame->tf_eip == (int)doreti_iret + setidt_disp) { 607 frame->tf_eip = (int)doreti_iret_fault + 608 setidt_disp; 609 return; 610 } 611 if (type == T_STKFLT) 612 break; 613 614 if (frame->tf_eip == (int)doreti_popl_ds + 615 setidt_disp) { 616 frame->tf_eip = (int)doreti_popl_ds_fault + 617 setidt_disp; 618 return; 619 } 620 if (frame->tf_eip == (int)doreti_popl_es + 621 setidt_disp) { 622 frame->tf_eip = (int)doreti_popl_es_fault + 623 setidt_disp; 624 return; 625 } 626 if (frame->tf_eip == (int)doreti_popl_fs + 627 setidt_disp) { 628 frame->tf_eip = (int)doreti_popl_fs_fault + 629 setidt_disp; 630 return; 631 } 632 if (curpcb->pcb_onfault != NULL) { 633 frame->tf_eip = (int)curpcb->pcb_onfault; 634 return; 635 } 636 break; 637 638 case T_TSSFLT: 639 /* 640 * PSL_NT can be set in user mode and isn't cleared 641 * automatically when the kernel is entered. This 642 * causes a TSS fault when the kernel attempts to 643 * `iret' because the TSS link is uninitialized. We 644 * want to get this fault so that we can fix the 645 * problem here and not every time the kernel is 646 * entered. 647 */ 648 if (frame->tf_eflags & PSL_NT) { 649 frame->tf_eflags &= ~PSL_NT; 650 return; 651 } 652 break; 653 654 case T_TRCTRAP: /* debug exception */ 655 kernel_trctrap: 656 /* Clear any pending debug events. */ 657 dr6 = rdr6(); 658 load_dr6(0); 659 660 /* 661 * Ignore debug register exceptions due to 662 * accesses in the user's address space, which 663 * can happen under several conditions such as 664 * if a user sets a watchpoint on a buffer and 665 * then passes that buffer to a system call. 666 * We still want to get TRCTRAPS for addresses 667 * in kernel space because that is useful when 668 * debugging the kernel. 669 */ 670 if (user_dbreg_trap(dr6) && 671 !(curpcb->pcb_flags & PCB_VM86CALL)) 672 return; 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) + setidt_disp || 698 frame->tf_eip == (uintptr_t)IDTVEC(bpt) + 699 setidt_disp || 700 frame->tf_eip == (uintptr_t)IDTVEC(dbg) + 701 setidt_disp) 702 return; 703 /* 704 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 705 */ 706 case T_BPTFLT: 707 /* 708 * If KDB is enabled, let it handle the debugger trap. 709 * Otherwise, debugger traps "can't happen". 710 */ 711 #ifdef KDB 712 if (kdb_trap(type, dr6, frame)) 713 return; 714 #endif 715 break; 716 717 #ifdef DEV_ISA 718 case T_NMI: 719 #ifdef POWERFAIL_NMI 720 if (time_second - lastalert > 10) { 721 log(LOG_WARNING, "NMI: power fail\n"); 722 sysbeep(880, hz); 723 lastalert = time_second; 724 } 725 return; 726 #else /* !POWERFAIL_NMI */ 727 nmi_handle_intr(type, frame); 728 return; 729 #endif /* POWERFAIL_NMI */ 730 #endif /* DEV_ISA */ 731 } 732 733 trap_fatal(frame, eva); 734 return; 735 } 736 737 /* Translate fault for emulators (e.g. Linux) */ 738 if (*p->p_sysent->sv_transtrap != NULL) 739 signo = (*p->p_sysent->sv_transtrap)(signo, type); 740 741 ksiginfo_init_trap(&ksi); 742 ksi.ksi_signo = signo; 743 ksi.ksi_code = ucode; 744 ksi.ksi_addr = (void *)addr; 745 ksi.ksi_trapno = type; 746 if (uprintf_signal) { 747 uprintf("pid %d comm %s: signal %d err %x code %d type %d " 748 "addr 0x%x ss 0x%04x esp 0x%08x cs 0x%04x eip 0x%08x " 749 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n", 750 p->p_pid, p->p_comm, signo, frame->tf_err, ucode, type, 751 addr, frame->tf_ss, frame->tf_esp, frame->tf_cs, 752 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 /* 766 * Clear any pending debug exceptions after allowing a 767 * debugger to read DR6 while stopped in trapsignal(). 768 */ 769 if (type == T_TRCTRAP) 770 load_dr6(0); 771 user: 772 userret(td, frame); 773 KASSERT(PCB_USER_FPU(td->td_pcb), 774 ("Return from trap with kernel FPU ctx leaked")); 775 } 776 777 static int 778 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva) 779 { 780 struct thread *td; 781 struct proc *p; 782 vm_offset_t va; 783 vm_map_t map; 784 int rv; 785 vm_prot_t ftype; 786 787 td = curthread; 788 p = td->td_proc; 789 790 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) { 791 /* 792 * Due to both processor errata and lazy TLB invalidation when 793 * access restrictions are removed from virtual pages, memory 794 * accesses that are allowed by the physical mapping layer may 795 * nonetheless cause one spurious page fault per virtual page. 796 * When the thread is executing a "no faulting" section that 797 * is bracketed by vm_fault_{disable,enable}_pagefaults(), 798 * every page fault is treated as a spurious page fault, 799 * unless it accesses the same virtual address as the most 800 * recent page fault within the same "no faulting" section. 801 */ 802 if (td->td_md.md_spurflt_addr != eva || 803 (td->td_pflags & TDP_RESETSPUR) != 0) { 804 /* 805 * Do nothing to the TLB. A stale TLB entry is 806 * flushed automatically by a page fault. 807 */ 808 td->td_md.md_spurflt_addr = eva; 809 td->td_pflags &= ~TDP_RESETSPUR; 810 return (0); 811 } 812 } else { 813 /* 814 * If we get a page fault while in a critical section, then 815 * it is most likely a fatal kernel page fault. The kernel 816 * is already going to panic trying to get a sleep lock to 817 * do the VM lookup, so just consider it a fatal trap so the 818 * kernel can print out a useful trap message and even get 819 * to the debugger. 820 * 821 * If we get a page fault while holding a non-sleepable 822 * lock, then it is most likely a fatal kernel page fault. 823 * If WITNESS is enabled, then it's going to whine about 824 * bogus LORs with various VM locks, so just skip to the 825 * fatal trap handling directly. 826 */ 827 if (td->td_critnest != 0 || 828 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL, 829 "Kernel page fault") != 0) { 830 trap_fatal(frame, eva); 831 return (-1); 832 } 833 } 834 va = trunc_page(eva); 835 if (va >= PMAP_TRM_MIN_ADDRESS) { 836 /* 837 * Don't allow user-mode faults in kernel address space. 838 * An exception: if the faulting address is the invalid 839 * instruction entry in the IDT, then the Intel Pentium 840 * F00F bug workaround was triggered, and we need to 841 * treat it is as an illegal instruction, and not a page 842 * fault. 843 */ 844 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 845 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) 846 return (-2); 847 #endif 848 if (usermode) 849 return (SIGSEGV); 850 trap_fatal(frame, eva); 851 return (-1); 852 } else { 853 map = usermode ? &p->p_vmspace->vm_map : kernel_map; 854 855 /* 856 * Kernel cannot access a user-space address directly 857 * because user pages are not mapped. Also, page 858 * faults must not be caused during the interrupts. 859 */ 860 if (!usermode && td->td_intr_nesting_level != 0) { 861 trap_fatal(frame, eva); 862 return (-1); 863 } 864 } 865 866 /* 867 * If the trap was caused by errant bits in the PTE then panic. 868 */ 869 if (frame->tf_err & PGEX_RSV) { 870 trap_fatal(frame, eva); 871 return (-1); 872 } 873 874 /* 875 * PGEX_I is defined only if the execute disable bit capability is 876 * supported and enabled. 877 */ 878 if (frame->tf_err & PGEX_W) 879 ftype = VM_PROT_WRITE; 880 #if defined(PAE) || defined(PAE_TABLES) 881 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 882 ftype = VM_PROT_EXECUTE; 883 #endif 884 else 885 ftype = VM_PROT_READ; 886 887 /* Fault in the page. */ 888 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 889 if (rv == KERN_SUCCESS) { 890 #ifdef HWPMC_HOOKS 891 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 892 PMC_SOFT_CALL_TF( , , page_fault, all, frame); 893 if (ftype == VM_PROT_READ) 894 PMC_SOFT_CALL_TF( , , page_fault, read, 895 frame); 896 else 897 PMC_SOFT_CALL_TF( , , page_fault, write, 898 frame); 899 } 900 #endif 901 return (0); 902 } 903 if (!usermode) { 904 if (td->td_intr_nesting_level == 0 && 905 curpcb->pcb_onfault != NULL) { 906 frame->tf_eip = (int)curpcb->pcb_onfault; 907 return (0); 908 } 909 trap_fatal(frame, eva); 910 return (-1); 911 } 912 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 913 } 914 915 static void 916 trap_fatal(frame, eva) 917 struct trapframe *frame; 918 vm_offset_t eva; 919 { 920 int code, ss, esp; 921 u_int type; 922 struct soft_segment_descriptor softseg; 923 #ifdef KDB 924 bool handled; 925 #endif 926 927 code = frame->tf_err; 928 type = frame->tf_trapno; 929 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 930 931 printf("\n\nFatal trap %d: %s while in %s mode\n", type, trap_msg(type), 932 frame->tf_eflags & PSL_VM ? "vm86" : 933 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 934 #ifdef SMP 935 /* two separate prints in case of a trap on an unmapped page */ 936 printf("cpuid = %d; ", PCPU_GET(cpuid)); 937 printf("apic id = %02x\n", PCPU_GET(apic_id)); 938 #endif 939 if (type == T_PAGEFLT) { 940 printf("fault virtual address = 0x%x\n", eva); 941 printf("fault code = %s %s%s, %s\n", 942 code & PGEX_U ? "user" : "supervisor", 943 code & PGEX_W ? "write" : "read", 944 #if defined(PAE) || defined(PAE_TABLES) 945 pg_nx != 0 ? 946 (code & PGEX_I ? " instruction" : " data") : 947 #endif 948 "", 949 code & PGEX_RSV ? "reserved bits in PTE" : 950 code & PGEX_P ? "protection violation" : "page not present"); 951 } else { 952 printf("error code = %#x\n", code); 953 } 954 printf("instruction pointer = 0x%x:0x%x\n", 955 frame->tf_cs & 0xffff, frame->tf_eip); 956 if (TF_HAS_STACKREGS(frame)) { 957 ss = frame->tf_ss & 0xffff; 958 esp = frame->tf_esp; 959 } else { 960 ss = GSEL(GDATA_SEL, SEL_KPL); 961 esp = (int)&frame->tf_esp; 962 } 963 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 964 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 965 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 966 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 967 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 968 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 969 softseg.ssd_gran); 970 printf("processor eflags = "); 971 if (frame->tf_eflags & PSL_T) 972 printf("trace trap, "); 973 if (frame->tf_eflags & PSL_I) 974 printf("interrupt enabled, "); 975 if (frame->tf_eflags & PSL_NT) 976 printf("nested task, "); 977 if (frame->tf_eflags & PSL_RF) 978 printf("resume, "); 979 if (frame->tf_eflags & PSL_VM) 980 printf("vm86, "); 981 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 982 printf("current process = %d (%s)\n", 983 curproc->p_pid, curthread->td_name); 984 985 #ifdef KDB 986 if (debugger_on_panic) { 987 kdb_why = KDB_WHY_TRAP; 988 frame->tf_err = eva; /* smuggle fault address to ddb */ 989 handled = kdb_trap(type, 0, frame); 990 frame->tf_err = code; /* restore error code */ 991 kdb_why = KDB_WHY_UNSET; 992 if (handled) 993 return; 994 } 995 #endif 996 printf("trap number = %d\n", type); 997 if (trap_msg(type) != NULL) 998 panic("%s", trap_msg(type)); 999 else 1000 panic("unknown/reserved trap"); 1001 } 1002 1003 /* 1004 * Double fault handler. Called when a fault occurs while writing 1005 * a frame for a trap/exception onto the stack. This usually occurs 1006 * when the stack overflows (such is the case with infinite recursion, 1007 * for example). 1008 * 1009 * XXX Note that the current PTD gets replaced by IdlePTD when the 1010 * task switch occurs. This means that the stack that was active at 1011 * the time of the double fault is not available at <kstack> unless 1012 * the machine was idle when the double fault occurred. The downside 1013 * of this is that "trace <ebp>" in ddb won't work. 1014 */ 1015 void 1016 dblfault_handler(void) 1017 { 1018 #ifdef KDTRACE_HOOKS 1019 if (dtrace_doubletrap_func != NULL) 1020 (*dtrace_doubletrap_func)(); 1021 #endif 1022 printf("\nFatal double fault:\n"); 1023 printf("eip = 0x%x\n", PCPU_GET(common_tssp)->tss_eip); 1024 printf("esp = 0x%x\n", PCPU_GET(common_tssp)->tss_esp); 1025 printf("ebp = 0x%x\n", PCPU_GET(common_tssp)->tss_ebp); 1026 #ifdef SMP 1027 /* two separate prints in case of a trap on an unmapped page */ 1028 printf("cpuid = %d; ", PCPU_GET(cpuid)); 1029 printf("apic id = %02x\n", PCPU_GET(apic_id)); 1030 #endif 1031 panic("double fault"); 1032 } 1033 1034 int 1035 cpu_fetch_syscall_args(struct thread *td) 1036 { 1037 struct proc *p; 1038 struct trapframe *frame; 1039 struct syscall_args *sa; 1040 caddr_t params; 1041 long tmp; 1042 int error; 1043 #ifdef COMPAT_43 1044 u_int32_t eip; 1045 int cs; 1046 #endif 1047 1048 p = td->td_proc; 1049 frame = td->td_frame; 1050 sa = &td->td_sa; 1051 1052 #ifdef COMPAT_43 1053 if (__predict_false(frame->tf_cs == 7 && frame->tf_eip == 2)) { 1054 /* 1055 * In lcall $7,$0 after int $0x80. Convert the user 1056 * frame to what it would be for a direct int 0x80 instead 1057 * of lcall $7,$0, by popping the lcall return address. 1058 */ 1059 error = fueword32((void *)frame->tf_esp, &eip); 1060 if (error == -1) 1061 return (EFAULT); 1062 cs = fuword16((void *)(frame->tf_esp + sizeof(u_int32_t))); 1063 if (cs == -1) 1064 return (EFAULT); 1065 1066 /* 1067 * Unwind in-kernel frame after all stack frame pieces 1068 * were successfully read. 1069 */ 1070 frame->tf_eip = eip; 1071 frame->tf_cs = cs; 1072 frame->tf_esp += 2 * sizeof(u_int32_t); 1073 frame->tf_err = 7; /* size of lcall $7,$0 */ 1074 } 1075 #endif 1076 1077 sa->code = frame->tf_eax; 1078 params = (caddr_t)frame->tf_esp + sizeof(uint32_t); 1079 1080 /* 1081 * Need to check if this is a 32 bit or 64 bit syscall. 1082 */ 1083 if (sa->code == SYS_syscall) { 1084 /* 1085 * Code is first argument, followed by actual args. 1086 */ 1087 error = fueword(params, &tmp); 1088 if (error == -1) 1089 return (EFAULT); 1090 sa->code = tmp; 1091 params += sizeof(uint32_t); 1092 } else if (sa->code == SYS___syscall) { 1093 /* 1094 * Like syscall, but code is a quad, so as to maintain 1095 * quad alignment for the rest of the arguments. 1096 */ 1097 error = fueword(params, &tmp); 1098 if (error == -1) 1099 return (EFAULT); 1100 sa->code = tmp; 1101 params += sizeof(quad_t); 1102 } 1103 1104 if (p->p_sysent->sv_mask) 1105 sa->code &= p->p_sysent->sv_mask; 1106 if (sa->code >= p->p_sysent->sv_size) 1107 sa->callp = &p->p_sysent->sv_table[0]; 1108 else 1109 sa->callp = &p->p_sysent->sv_table[sa->code]; 1110 sa->narg = sa->callp->sy_narg; 1111 1112 if (params != NULL && sa->narg != 0) 1113 error = copyin(params, (caddr_t)sa->args, 1114 (u_int)(sa->narg * sizeof(uint32_t))); 1115 else 1116 error = 0; 1117 1118 if (error == 0) { 1119 td->td_retval[0] = 0; 1120 td->td_retval[1] = frame->tf_edx; 1121 } 1122 1123 return (error); 1124 } 1125 1126 #include "../../kern/subr_syscall.c" 1127 1128 /* 1129 * syscall - system call request C handler. A system call is 1130 * essentially treated as a trap by reusing the frame layout. 1131 */ 1132 void 1133 syscall(struct trapframe *frame) 1134 { 1135 struct thread *td; 1136 register_t orig_tf_eflags; 1137 int error; 1138 ksiginfo_t ksi; 1139 1140 #ifdef DIAGNOSTIC 1141 if (!(TRAPF_USERMODE(frame) && 1142 (curpcb->pcb_flags & PCB_VM86CALL) == 0)) { 1143 panic("syscall"); 1144 /* NOT REACHED */ 1145 } 1146 #endif 1147 orig_tf_eflags = frame->tf_eflags; 1148 1149 td = curthread; 1150 td->td_frame = frame; 1151 1152 error = syscallenter(td); 1153 1154 /* 1155 * Traced syscall. 1156 */ 1157 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) { 1158 frame->tf_eflags &= ~PSL_T; 1159 ksiginfo_init_trap(&ksi); 1160 ksi.ksi_signo = SIGTRAP; 1161 ksi.ksi_code = TRAP_TRACE; 1162 ksi.ksi_addr = (void *)frame->tf_eip; 1163 trapsignal(td, &ksi); 1164 } 1165 1166 KASSERT(PCB_USER_FPU(td->td_pcb), 1167 ("System call %s returning with kernel FPU ctx leaked", 1168 syscallname(td->td_proc, td->td_sa.code))); 1169 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td), 1170 ("System call %s returning with mangled pcb_save", 1171 syscallname(td->td_proc, td->td_sa.code))); 1172 1173 syscallret(td, error); 1174 } 1175