trap.c revision 287146
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/10.1/sys/amd64/amd64/trap.c 287146 2015-08-25 20:48:58Z delphij $"); 42 43/* 44 * AMD64 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 54#include <sys/param.h> 55#include <sys/bus.h> 56#include <sys/systm.h> 57#include <sys/proc.h> 58#include <sys/pioctl.h> 59#include <sys/ptrace.h> 60#include <sys/kdb.h> 61#include <sys/kernel.h> 62#include <sys/ktr.h> 63#include <sys/lock.h> 64#include <sys/mutex.h> 65#include <sys/resourcevar.h> 66#include <sys/signalvar.h> 67#include <sys/syscall.h> 68#include <sys/sysctl.h> 69#include <sys/sysent.h> 70#include <sys/uio.h> 71#include <sys/vmmeter.h> 72#ifdef HWPMC_HOOKS 73#include <sys/pmckern.h> 74PMC_SOFT_DEFINE( , , page_fault, all); 75PMC_SOFT_DEFINE( , , page_fault, read); 76PMC_SOFT_DEFINE( , , page_fault, write); 77#endif 78 79#include <vm/vm.h> 80#include <vm/vm_param.h> 81#include <vm/pmap.h> 82#include <vm/vm_kern.h> 83#include <vm/vm_map.h> 84#include <vm/vm_page.h> 85#include <vm/vm_extern.h> 86 87#include <machine/cpu.h> 88#include <machine/intr_machdep.h> 89#include <x86/mca.h> 90#include <machine/md_var.h> 91#include <machine/pcb.h> 92#ifdef SMP 93#include <machine/smp.h> 94#endif 95#include <machine/tss.h> 96 97#ifdef KDTRACE_HOOKS 98#include <sys/dtrace_bsd.h> 99#endif 100 101extern void trap(struct trapframe *frame); 102extern void syscall(struct trapframe *frame); 103void dblfault_handler(struct trapframe *frame); 104 105static int trap_pfault(struct trapframe *, int); 106static void trap_fatal(struct trapframe *, vm_offset_t); 107 108#define MAX_TRAP_MSG 32 109static char *trap_msg[] = { 110 "", /* 0 unused */ 111 "privileged instruction fault", /* 1 T_PRIVINFLT */ 112 "", /* 2 unused */ 113 "breakpoint instruction fault", /* 3 T_BPTFLT */ 114 "", /* 4 unused */ 115 "", /* 5 unused */ 116 "arithmetic trap", /* 6 T_ARITHTRAP */ 117 "", /* 7 unused */ 118 "", /* 8 unused */ 119 "general protection fault", /* 9 T_PROTFLT */ 120 "trace trap", /* 10 T_TRCTRAP */ 121 "", /* 11 unused */ 122 "page fault", /* 12 T_PAGEFLT */ 123 "", /* 13 unused */ 124 "alignment fault", /* 14 T_ALIGNFLT */ 125 "", /* 15 unused */ 126 "", /* 16 unused */ 127 "", /* 17 unused */ 128 "integer divide fault", /* 18 T_DIVIDE */ 129 "non-maskable interrupt trap", /* 19 T_NMI */ 130 "overflow trap", /* 20 T_OFLOW */ 131 "FPU bounds check fault", /* 21 T_BOUND */ 132 "FPU device not available", /* 22 T_DNA */ 133 "double fault", /* 23 T_DOUBLEFLT */ 134 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 135 "invalid TSS fault", /* 25 T_TSSFLT */ 136 "segment not present fault", /* 26 T_SEGNPFLT */ 137 "stack fault", /* 27 T_STKFLT */ 138 "machine check trap", /* 28 T_MCHK */ 139 "SIMD floating-point exception", /* 29 T_XMMFLT */ 140 "reserved (unknown) fault", /* 30 T_RESERVED */ 141 "", /* 31 unused (reserved) */ 142 "DTrace pid return trap", /* 32 T_DTRACE_RET */ 143}; 144 145#ifdef KDB 146static int kdb_on_nmi = 1; 147SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW, 148 &kdb_on_nmi, 0, "Go to KDB on NMI"); 149TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi); 150#endif 151static int panic_on_nmi = 1; 152SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 153 &panic_on_nmi, 0, "Panic on NMI"); 154TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi); 155static int prot_fault_translation; 156SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW, 157 &prot_fault_translation, 0, 158 "Select signal to deliver on protection fault"); 159static int uprintf_signal; 160SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW, 161 &uprintf_signal, 0, 162 "Print debugging information on trap signal to ctty"); 163 164/* 165 * Exception, fault, and trap interface to the FreeBSD kernel. 166 * This common code is called from assembly language IDT gate entry 167 * routines that prepare a suitable stack frame, and restore this 168 * frame after the exception has been processed. 169 */ 170 171void 172trap(struct trapframe *frame) 173{ 174#ifdef KDTRACE_HOOKS 175 struct reg regs; 176#endif 177 struct thread *td = curthread; 178 struct proc *p = td->td_proc; 179 int i = 0, ucode = 0, code; 180 u_int type; 181 register_t addr = 0; 182 ksiginfo_t ksi; 183 184 PCPU_INC(cnt.v_trap); 185 type = frame->tf_trapno; 186 187#ifdef SMP 188 /* Handler for NMI IPIs used for stopping CPUs. */ 189 if (type == T_NMI) { 190 if (ipi_nmi_handler() == 0) 191 goto out; 192 } 193#endif /* SMP */ 194 195#ifdef KDB 196 if (kdb_active) { 197 kdb_reenter(); 198 goto out; 199 } 200#endif 201 202 if (type == T_RESERVED) { 203 trap_fatal(frame, 0); 204 goto out; 205 } 206 207#ifdef HWPMC_HOOKS 208 /* 209 * CPU PMCs interrupt using an NMI. If the PMC module is 210 * active, pass the 'rip' value to the PMC module's interrupt 211 * handler. A return value of '1' from the handler means that 212 * the NMI was handled by it and we can return immediately. 213 */ 214 if (type == T_NMI && pmc_intr && 215 (*pmc_intr)(PCPU_GET(cpuid), frame)) 216 goto out; 217#endif 218 219 if (type == T_MCHK) { 220 mca_intr(); 221 goto out; 222 } 223 224#ifdef KDTRACE_HOOKS 225 /* 226 * A trap can occur while DTrace executes a probe. Before 227 * executing the probe, DTrace blocks re-scheduling and sets 228 * a flag in its per-cpu flags to indicate that it doesn't 229 * want to fault. On returning from the probe, the no-fault 230 * flag is cleared and finally re-scheduling is enabled. 231 */ 232 if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type)) 233 goto out; 234#endif 235 236 if ((frame->tf_rflags & PSL_I) == 0) { 237 /* 238 * Buggy application or kernel code has disabled 239 * interrupts and then trapped. Enabling interrupts 240 * now is wrong, but it is better than running with 241 * interrupts disabled until they are accidentally 242 * enabled later. 243 */ 244 if (ISPL(frame->tf_cs) == SEL_UPL) 245 uprintf( 246 "pid %ld (%s): trap %d with interrupts disabled\n", 247 (long)curproc->p_pid, curthread->td_name, type); 248 else if (type != T_NMI && type != T_BPTFLT && 249 type != T_TRCTRAP) { 250 /* 251 * XXX not quite right, since this may be for a 252 * multiple fault in user mode. 253 */ 254 printf("kernel trap %d with interrupts disabled\n", 255 type); 256 257 /* 258 * We shouldn't enable interrupts while holding a 259 * spin lock. 260 */ 261 if (td->td_md.md_spinlock_count == 0) 262 enable_intr(); 263 } 264 } 265 266 code = frame->tf_err; 267 268 if (ISPL(frame->tf_cs) == SEL_UPL) { 269 /* user trap */ 270 271 td->td_pticks = 0; 272 td->td_frame = frame; 273 addr = frame->tf_rip; 274 if (td->td_ucred != p->p_ucred) 275 cred_update_thread(td); 276 277 switch (type) { 278 case T_PRIVINFLT: /* privileged instruction fault */ 279 i = SIGILL; 280 ucode = ILL_PRVOPC; 281 break; 282 283 case T_BPTFLT: /* bpt instruction fault */ 284 case T_TRCTRAP: /* trace trap */ 285 enable_intr(); 286#ifdef KDTRACE_HOOKS 287 if (type == T_BPTFLT) { 288 fill_frame_regs(frame, ®s); 289 if (dtrace_pid_probe_ptr != NULL && 290 dtrace_pid_probe_ptr(®s) == 0) 291 goto out; 292 } 293#endif 294 frame->tf_rflags &= ~PSL_T; 295 i = SIGTRAP; 296 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 297 break; 298 299 case T_ARITHTRAP: /* arithmetic trap */ 300 ucode = fputrap_x87(); 301 if (ucode == -1) 302 goto userout; 303 i = SIGFPE; 304 break; 305 306 case T_PROTFLT: /* general protection fault */ 307 i = SIGBUS; 308 ucode = BUS_OBJERR; 309 break; 310 case T_STKFLT: /* stack fault */ 311 case T_SEGNPFLT: /* segment not present fault */ 312 i = SIGBUS; 313 ucode = BUS_ADRERR; 314 break; 315 case T_TSSFLT: /* invalid TSS fault */ 316 i = SIGBUS; 317 ucode = BUS_OBJERR; 318 break; 319 case T_ALIGNFLT: 320 i = SIGBUS; 321 ucode = BUS_ADRALN; 322 break; 323 case T_DOUBLEFLT: /* double fault */ 324 default: 325 i = SIGBUS; 326 ucode = BUS_OBJERR; 327 break; 328 329 case T_PAGEFLT: /* page fault */ 330 addr = frame->tf_addr; 331 i = trap_pfault(frame, TRUE); 332 if (i == -1) 333 goto userout; 334 if (i == 0) 335 goto user; 336 337 if (i == SIGSEGV) 338 ucode = SEGV_MAPERR; 339 else { 340 if (prot_fault_translation == 0) { 341 /* 342 * Autodetect. 343 * This check also covers the images 344 * without the ABI-tag ELF note. 345 */ 346 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD 347 && p->p_osrel >= P_OSREL_SIGSEGV) { 348 i = SIGSEGV; 349 ucode = SEGV_ACCERR; 350 } else { 351 i = SIGBUS; 352 ucode = BUS_PAGE_FAULT; 353 } 354 } else if (prot_fault_translation == 1) { 355 /* 356 * Always compat mode. 357 */ 358 i = SIGBUS; 359 ucode = BUS_PAGE_FAULT; 360 } else { 361 /* 362 * Always SIGSEGV mode. 363 */ 364 i = SIGSEGV; 365 ucode = SEGV_ACCERR; 366 } 367 } 368 break; 369 370 case T_DIVIDE: /* integer divide fault */ 371 ucode = FPE_INTDIV; 372 i = SIGFPE; 373 break; 374 375#ifdef DEV_ISA 376 case T_NMI: 377 /* machine/parity/power fail/"kitchen sink" faults */ 378 if (isa_nmi(code) == 0) { 379#ifdef KDB 380 /* 381 * NMI can be hooked up to a pushbutton 382 * for debugging. 383 */ 384 if (kdb_on_nmi) { 385 printf ("NMI ... going to debugger\n"); 386 kdb_trap(type, 0, frame); 387 } 388#endif /* KDB */ 389 goto userout; 390 } else if (panic_on_nmi) 391 panic("NMI indicates hardware failure"); 392 break; 393#endif /* DEV_ISA */ 394 395 case T_OFLOW: /* integer overflow fault */ 396 ucode = FPE_INTOVF; 397 i = SIGFPE; 398 break; 399 400 case T_BOUND: /* bounds check fault */ 401 ucode = FPE_FLTSUB; 402 i = SIGFPE; 403 break; 404 405 case T_DNA: 406 /* transparent fault (due to context switch "late") */ 407 KASSERT(PCB_USER_FPU(td->td_pcb), 408 ("kernel FPU ctx has leaked")); 409 fpudna(); 410 goto userout; 411 412 case T_FPOPFLT: /* FPU operand fetch fault */ 413 ucode = ILL_COPROC; 414 i = SIGILL; 415 break; 416 417 case T_XMMFLT: /* SIMD floating-point exception */ 418 ucode = fputrap_sse(); 419 if (ucode == -1) 420 goto userout; 421 i = SIGFPE; 422 break; 423#ifdef KDTRACE_HOOKS 424 case T_DTRACE_RET: 425 enable_intr(); 426 fill_frame_regs(frame, ®s); 427 if (dtrace_return_probe_ptr != NULL && 428 dtrace_return_probe_ptr(®s) == 0) 429 goto out; 430 break; 431#endif 432 } 433 } else { 434 /* kernel trap */ 435 436 KASSERT(cold || td->td_ucred != NULL, 437 ("kernel trap doesn't have ucred")); 438 switch (type) { 439 case T_PAGEFLT: /* page fault */ 440 (void) trap_pfault(frame, FALSE); 441 goto out; 442 443 case T_DNA: 444 KASSERT(!PCB_USER_FPU(td->td_pcb), 445 ("Unregistered use of FPU in kernel")); 446 fpudna(); 447 goto out; 448 449 case T_ARITHTRAP: /* arithmetic trap */ 450 case T_XMMFLT: /* SIMD floating-point exception */ 451 case T_FPOPFLT: /* FPU operand fetch fault */ 452 /* 453 * XXXKIB for now disable any FPU traps in kernel 454 * handler registration seems to be overkill 455 */ 456 trap_fatal(frame, 0); 457 goto out; 458 459 case T_STKFLT: /* stack fault */ 460 case T_PROTFLT: /* general protection fault */ 461 case T_SEGNPFLT: /* segment not present fault */ 462 if (td->td_intr_nesting_level != 0) 463 break; 464 465 /* 466 * Invalid segment selectors and out of bounds 467 * %rip's and %rsp's can be set up in user mode. 468 * This causes a fault in kernel mode when the 469 * kernel tries to return to user mode. We want 470 * to get this fault so that we can fix the 471 * problem here and not have to check all the 472 * selectors and pointers when the user changes 473 * them. 474 */ 475 if (frame->tf_rip == (long)doreti_iret) { 476 frame->tf_rip = (long)doreti_iret_fault; 477 goto out; 478 } 479 if (frame->tf_rip == (long)ld_ds) { 480 frame->tf_rip = (long)ds_load_fault; 481 goto out; 482 } 483 if (frame->tf_rip == (long)ld_es) { 484 frame->tf_rip = (long)es_load_fault; 485 goto out; 486 } 487 if (frame->tf_rip == (long)ld_fs) { 488 frame->tf_rip = (long)fs_load_fault; 489 goto out; 490 } 491 if (frame->tf_rip == (long)ld_gs) { 492 frame->tf_rip = (long)gs_load_fault; 493 goto out; 494 } 495 if (frame->tf_rip == (long)ld_gsbase) { 496 frame->tf_rip = (long)gsbase_load_fault; 497 goto out; 498 } 499 if (frame->tf_rip == (long)ld_fsbase) { 500 frame->tf_rip = (long)fsbase_load_fault; 501 goto out; 502 } 503 if (curpcb->pcb_onfault != NULL) { 504 frame->tf_rip = (long)curpcb->pcb_onfault; 505 goto out; 506 } 507 break; 508 509 case T_TSSFLT: 510 /* 511 * PSL_NT can be set in user mode and isn't cleared 512 * automatically when the kernel is entered. This 513 * causes a TSS fault when the kernel attempts to 514 * `iret' because the TSS link is uninitialized. We 515 * want to get this fault so that we can fix the 516 * problem here and not every time the kernel is 517 * entered. 518 */ 519 if (frame->tf_rflags & PSL_NT) { 520 frame->tf_rflags &= ~PSL_NT; 521 goto out; 522 } 523 break; 524 525 case T_TRCTRAP: /* trace trap */ 526 /* 527 * Ignore debug register trace traps due to 528 * accesses in the user's address space, which 529 * can happen under several conditions such as 530 * if a user sets a watchpoint on a buffer and 531 * then passes that buffer to a system call. 532 * We still want to get TRCTRAPS for addresses 533 * in kernel space because that is useful when 534 * debugging the kernel. 535 */ 536 if (user_dbreg_trap()) { 537 /* 538 * Reset breakpoint bits because the 539 * processor doesn't 540 */ 541 /* XXX check upper bits here */ 542 load_dr6(rdr6() & 0xfffffff0); 543 goto out; 544 } 545 /* 546 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 547 */ 548 case T_BPTFLT: 549 /* 550 * If KDB is enabled, let it handle the debugger trap. 551 * Otherwise, debugger traps "can't happen". 552 */ 553#ifdef KDB 554 if (kdb_trap(type, 0, frame)) 555 goto out; 556#endif 557 break; 558 559#ifdef DEV_ISA 560 case T_NMI: 561 /* machine/parity/power fail/"kitchen sink" faults */ 562 if (isa_nmi(code) == 0) { 563#ifdef KDB 564 /* 565 * NMI can be hooked up to a pushbutton 566 * for debugging. 567 */ 568 if (kdb_on_nmi) { 569 printf ("NMI ... going to debugger\n"); 570 kdb_trap(type, 0, frame); 571 } 572#endif /* KDB */ 573 goto out; 574 } else if (panic_on_nmi == 0) 575 goto out; 576 /* FALLTHROUGH */ 577#endif /* DEV_ISA */ 578 } 579 580 trap_fatal(frame, 0); 581 goto out; 582 } 583 584 /* Translate fault for emulators (e.g. Linux) */ 585 if (*p->p_sysent->sv_transtrap) 586 i = (*p->p_sysent->sv_transtrap)(i, type); 587 588 ksiginfo_init_trap(&ksi); 589 ksi.ksi_signo = i; 590 ksi.ksi_code = ucode; 591 ksi.ksi_trapno = type; 592 ksi.ksi_addr = (void *)addr; 593 if (uprintf_signal) { 594 uprintf("pid %d comm %s: signal %d err %lx code %d type %d " 595 "addr 0x%lx rsp 0x%lx rip 0x%lx " 596 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n", 597 p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr, 598 frame->tf_rsp, frame->tf_rip, 599 fubyte((void *)(frame->tf_rip + 0)), 600 fubyte((void *)(frame->tf_rip + 1)), 601 fubyte((void *)(frame->tf_rip + 2)), 602 fubyte((void *)(frame->tf_rip + 3)), 603 fubyte((void *)(frame->tf_rip + 4)), 604 fubyte((void *)(frame->tf_rip + 5)), 605 fubyte((void *)(frame->tf_rip + 6)), 606 fubyte((void *)(frame->tf_rip + 7))); 607 } 608 KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled")); 609 trapsignal(td, &ksi); 610 611user: 612 userret(td, frame); 613 KASSERT(PCB_USER_FPU(td->td_pcb), 614 ("Return from trap with kernel FPU ctx leaked")); 615userout: 616out: 617 return; 618} 619 620static int 621trap_pfault(frame, usermode) 622 struct trapframe *frame; 623 int usermode; 624{ 625 vm_offset_t va; 626 struct vmspace *vm; 627 vm_map_t map; 628 int rv = 0; 629 vm_prot_t ftype; 630 struct thread *td = curthread; 631 struct proc *p = td->td_proc; 632 vm_offset_t eva = frame->tf_addr; 633 634 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) { 635 /* 636 * Due to both processor errata and lazy TLB invalidation when 637 * access restrictions are removed from virtual pages, memory 638 * accesses that are allowed by the physical mapping layer may 639 * nonetheless cause one spurious page fault per virtual page. 640 * When the thread is executing a "no faulting" section that 641 * is bracketed by vm_fault_{disable,enable}_pagefaults(), 642 * every page fault is treated as a spurious page fault, 643 * unless it accesses the same virtual address as the most 644 * recent page fault within the same "no faulting" section. 645 */ 646 if (td->td_md.md_spurflt_addr != eva || 647 (td->td_pflags & TDP_RESETSPUR) != 0) { 648 /* 649 * Do nothing to the TLB. A stale TLB entry is 650 * flushed automatically by a page fault. 651 */ 652 td->td_md.md_spurflt_addr = eva; 653 td->td_pflags &= ~TDP_RESETSPUR; 654 return (0); 655 } 656 } else { 657 /* 658 * If we get a page fault while in a critical section, then 659 * it is most likely a fatal kernel page fault. The kernel 660 * is already going to panic trying to get a sleep lock to 661 * do the VM lookup, so just consider it a fatal trap so the 662 * kernel can print out a useful trap message and even get 663 * to the debugger. 664 * 665 * If we get a page fault while holding a non-sleepable 666 * lock, then it is most likely a fatal kernel page fault. 667 * If WITNESS is enabled, then it's going to whine about 668 * bogus LORs with various VM locks, so just skip to the 669 * fatal trap handling directly. 670 */ 671 if (td->td_critnest != 0 || 672 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL, 673 "Kernel page fault") != 0) { 674 trap_fatal(frame, eva); 675 return (-1); 676 } 677 } 678 va = trunc_page(eva); 679 if (va >= VM_MIN_KERNEL_ADDRESS) { 680 /* 681 * Don't allow user-mode faults in kernel address space. 682 */ 683 if (usermode) 684 goto nogo; 685 686 map = kernel_map; 687 } else { 688 /* 689 * This is a fault on non-kernel virtual memory. If either 690 * p or p->p_vmspace is NULL, then the fault is fatal. 691 */ 692 if (p == NULL || (vm = p->p_vmspace) == NULL) 693 goto nogo; 694 695 map = &vm->vm_map; 696 697 /* 698 * When accessing a usermode address, kernel must be 699 * ready to accept the page fault, and provide a 700 * handling routine. Since accessing the address 701 * without the handler is a bug, do not try to handle 702 * it normally, and panic immediately. 703 */ 704 if (!usermode && (td->td_intr_nesting_level != 0 || 705 curpcb->pcb_onfault == NULL)) { 706 trap_fatal(frame, eva); 707 return (-1); 708 } 709 } 710 711 /* 712 * If the trap was caused by errant bits in the PTE then panic. 713 */ 714 if (frame->tf_err & PGEX_RSV) { 715 trap_fatal(frame, eva); 716 return (-1); 717 } 718 719 /* 720 * PGEX_I is defined only if the execute disable bit capability is 721 * supported and enabled. 722 */ 723 if (frame->tf_err & PGEX_W) 724 ftype = VM_PROT_WRITE; 725 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 726 ftype = VM_PROT_EXECUTE; 727 else 728 ftype = VM_PROT_READ; 729 730 if (map != kernel_map) { 731 /* 732 * Keep swapout from messing with us during this 733 * critical time. 734 */ 735 PROC_LOCK(p); 736 ++p->p_lock; 737 PROC_UNLOCK(p); 738 739 /* Fault in the user page: */ 740 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 741 742 PROC_LOCK(p); 743 --p->p_lock; 744 PROC_UNLOCK(p); 745 } else { 746 /* 747 * Don't have to worry about process locking or stacks in the 748 * kernel. 749 */ 750 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 751 } 752 if (rv == KERN_SUCCESS) { 753#ifdef HWPMC_HOOKS 754 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 755 PMC_SOFT_CALL_TF( , , page_fault, all, frame); 756 if (ftype == VM_PROT_READ) 757 PMC_SOFT_CALL_TF( , , page_fault, read, 758 frame); 759 else 760 PMC_SOFT_CALL_TF( , , page_fault, write, 761 frame); 762 } 763#endif 764 return (0); 765 } 766nogo: 767 if (!usermode) { 768 if (td->td_intr_nesting_level == 0 && 769 curpcb->pcb_onfault != NULL) { 770 frame->tf_rip = (long)curpcb->pcb_onfault; 771 return (0); 772 } 773 if ((td->td_pflags & TDP_DEVMEMIO) != 0) { 774 KASSERT(curpcb->pcb_onfault != NULL, 775 ("/dev/mem without pcb_onfault")); 776 frame->tf_rip = (long)curpcb->pcb_onfault; 777 return (0); 778 } 779 trap_fatal(frame, eva); 780 return (-1); 781 } 782 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 783} 784 785static void 786trap_fatal(frame, eva) 787 struct trapframe *frame; 788 vm_offset_t eva; 789{ 790 int code, ss; 791 u_int type; 792 long esp; 793 struct soft_segment_descriptor softseg; 794 char *msg; 795 796 code = frame->tf_err; 797 type = frame->tf_trapno; 798 sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)], 799 &softseg); 800 801 if (type <= MAX_TRAP_MSG) 802 msg = trap_msg[type]; 803 else 804 msg = "UNKNOWN"; 805 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 806 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 807#ifdef SMP 808 /* two separate prints in case of a trap on an unmapped page */ 809 printf("cpuid = %d; ", PCPU_GET(cpuid)); 810 printf("apic id = %02x\n", PCPU_GET(apic_id)); 811#endif 812 if (type == T_PAGEFLT) { 813 printf("fault virtual address = 0x%lx\n", eva); 814 printf("fault code = %s %s %s%s, %s\n", 815 code & PGEX_U ? "user" : "supervisor", 816 code & PGEX_W ? "write" : "read", 817 code & PGEX_I ? "instruction" : "data", 818 code & PGEX_RSV ? " rsv" : "", 819 code & PGEX_P ? "protection violation" : "page not present"); 820 } 821 printf("instruction pointer = 0x%lx:0x%lx\n", 822 frame->tf_cs & 0xffff, frame->tf_rip); 823 if (ISPL(frame->tf_cs) == SEL_UPL) { 824 ss = frame->tf_ss & 0xffff; 825 esp = frame->tf_rsp; 826 } else { 827 ss = GSEL(GDATA_SEL, SEL_KPL); 828 esp = (long)&frame->tf_rsp; 829 } 830 printf("stack pointer = 0x%x:0x%lx\n", ss, esp); 831 printf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 832 printf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 833 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 834 printf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 835 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 836 softseg.ssd_gran); 837 printf("processor eflags = "); 838 if (frame->tf_rflags & PSL_T) 839 printf("trace trap, "); 840 if (frame->tf_rflags & PSL_I) 841 printf("interrupt enabled, "); 842 if (frame->tf_rflags & PSL_NT) 843 printf("nested task, "); 844 if (frame->tf_rflags & PSL_RF) 845 printf("resume, "); 846 printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 847 printf("current process = "); 848 if (curproc) { 849 printf("%lu (%s)\n", 850 (u_long)curproc->p_pid, curthread->td_name ? 851 curthread->td_name : ""); 852 } else { 853 printf("Idle\n"); 854 } 855 856#ifdef KDB 857 if (debugger_on_panic || kdb_active) 858 if (kdb_trap(type, 0, frame)) 859 return; 860#endif 861 printf("trap number = %d\n", type); 862 if (type <= MAX_TRAP_MSG) 863 panic("%s", trap_msg[type]); 864 else 865 panic("unknown/reserved trap"); 866} 867 868/* 869 * Double fault handler. Called when a fault occurs while writing 870 * a frame for a trap/exception onto the stack. This usually occurs 871 * when the stack overflows (such is the case with infinite recursion, 872 * for example). 873 */ 874void 875dblfault_handler(struct trapframe *frame) 876{ 877#ifdef KDTRACE_HOOKS 878 if (dtrace_doubletrap_func != NULL) 879 (*dtrace_doubletrap_func)(); 880#endif 881 printf("\nFatal double fault\n"); 882 printf("rip = 0x%lx\n", frame->tf_rip); 883 printf("rsp = 0x%lx\n", frame->tf_rsp); 884 printf("rbp = 0x%lx\n", frame->tf_rbp); 885#ifdef SMP 886 /* two separate prints in case of a trap on an unmapped page */ 887 printf("cpuid = %d; ", PCPU_GET(cpuid)); 888 printf("apic id = %02x\n", PCPU_GET(apic_id)); 889#endif 890 panic("double fault"); 891} 892 893int 894cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa) 895{ 896 struct proc *p; 897 struct trapframe *frame; 898 register_t *argp; 899 caddr_t params; 900 int reg, regcnt, error; 901 902 p = td->td_proc; 903 frame = td->td_frame; 904 reg = 0; 905 regcnt = 6; 906 907 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 908 sa->code = frame->tf_rax; 909 910 if (sa->code == SYS_syscall || sa->code == SYS___syscall) { 911 sa->code = frame->tf_rdi; 912 reg++; 913 regcnt--; 914 } 915 if (p->p_sysent->sv_mask) 916 sa->code &= p->p_sysent->sv_mask; 917 918 if (sa->code >= p->p_sysent->sv_size) 919 sa->callp = &p->p_sysent->sv_table[0]; 920 else 921 sa->callp = &p->p_sysent->sv_table[sa->code]; 922 923 sa->narg = sa->callp->sy_narg; 924 KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]), 925 ("Too many syscall arguments!")); 926 error = 0; 927 argp = &frame->tf_rdi; 928 argp += reg; 929 bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt); 930 if (sa->narg > regcnt) { 931 KASSERT(params != NULL, ("copyin args with no params!")); 932 error = copyin(params, &sa->args[regcnt], 933 (sa->narg - regcnt) * sizeof(sa->args[0])); 934 } 935 936 if (error == 0) { 937 td->td_retval[0] = 0; 938 td->td_retval[1] = frame->tf_rdx; 939 } 940 941 return (error); 942} 943 944#include "../../kern/subr_syscall.c" 945 946/* 947 * System call handler for native binaries. The trap frame is already 948 * set up by the assembler trampoline and a pointer to it is saved in 949 * td_frame. 950 */ 951void 952amd64_syscall(struct thread *td, int traced) 953{ 954 struct syscall_args sa; 955 int error; 956 ksiginfo_t ksi; 957 958#ifdef DIAGNOSTIC 959 if (ISPL(td->td_frame->tf_cs) != SEL_UPL) { 960 panic("syscall"); 961 /* NOT REACHED */ 962 } 963#endif 964 error = syscallenter(td, &sa); 965 966 /* 967 * Traced syscall. 968 */ 969 if (__predict_false(traced)) { 970 td->td_frame->tf_rflags &= ~PSL_T; 971 ksiginfo_init_trap(&ksi); 972 ksi.ksi_signo = SIGTRAP; 973 ksi.ksi_code = TRAP_TRACE; 974 ksi.ksi_addr = (void *)td->td_frame->tf_rip; 975 trapsignal(td, &ksi); 976 } 977 978 KASSERT(PCB_USER_FPU(td->td_pcb), 979 ("System call %s returing with kernel FPU ctx leaked", 980 syscallname(td->td_proc, sa.code))); 981 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td), 982 ("System call %s returning with mangled pcb_save", 983 syscallname(td->td_proc, sa.code))); 984 985 syscallret(td, error, &sa); 986 987 /* 988 * If the user-supplied value of %rip is not a canonical 989 * address, then some CPUs will trigger a ring 0 #GP during 990 * the sysret instruction. However, the fault handler would 991 * execute in ring 0 with the user's %gs and %rsp which would 992 * not be safe. Instead, use the full return path which 993 * catches the problem safely. 994 */ 995 if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS) 996 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 997} 998