1/*- 2 * Copyright (C) 1994, David Greenman 3 * Copyright (c) 1990, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * the University of Utah, and William Jolitz. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the University of 20 * California, Berkeley and its contributors. 21 * 4. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 38 */ 39 40#include <sys/cdefs.h> 41__FBSDID("$FreeBSD$"); 42 43/* 44 * 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> 99 100/* 101 * This is a hook which is initialised by the dtrace module 102 * to handle traps which might occur during DTrace probe 103 * execution. 104 */ 105dtrace_trap_func_t dtrace_trap_func; 106 107dtrace_doubletrap_func_t dtrace_doubletrap_func; 108 109/* 110 * This is a hook which is initialised by the systrace module 111 * when it is loaded. This keeps the DTrace syscall provider 112 * implementation opaque. 113 */ 114systrace_probe_func_t systrace_probe_func; 115 116/* 117 * These hooks are necessary for the pid, usdt and fasttrap providers. 118 */ 119dtrace_fasttrap_probe_ptr_t dtrace_fasttrap_probe_ptr; 120dtrace_pid_probe_ptr_t dtrace_pid_probe_ptr; 121dtrace_return_probe_ptr_t dtrace_return_probe_ptr; 122#endif 123 124extern void trap(struct trapframe *frame); 125extern void syscall(struct trapframe *frame); 126void dblfault_handler(struct trapframe *frame); 127 128static int trap_pfault(struct trapframe *, int); 129static void trap_fatal(struct trapframe *, vm_offset_t); 130 131#define MAX_TRAP_MSG 33 132static char *trap_msg[] = { 133 "", /* 0 unused */ 134 "privileged instruction fault", /* 1 T_PRIVINFLT */ 135 "", /* 2 unused */ 136 "breakpoint instruction fault", /* 3 T_BPTFLT */ 137 "", /* 4 unused */ 138 "", /* 5 unused */ 139 "arithmetic trap", /* 6 T_ARITHTRAP */ 140 "", /* 7 unused */ 141 "", /* 8 unused */ 142 "general protection fault", /* 9 T_PROTFLT */ 143 "trace trap", /* 10 T_TRCTRAP */ 144 "", /* 11 unused */ 145 "page fault", /* 12 T_PAGEFLT */ 146 "", /* 13 unused */ 147 "alignment fault", /* 14 T_ALIGNFLT */ 148 "", /* 15 unused */ 149 "", /* 16 unused */ 150 "", /* 17 unused */ 151 "integer divide fault", /* 18 T_DIVIDE */ 152 "non-maskable interrupt trap", /* 19 T_NMI */ 153 "overflow trap", /* 20 T_OFLOW */ 154 "FPU bounds check fault", /* 21 T_BOUND */ 155 "FPU device not available", /* 22 T_DNA */ 156 "double fault", /* 23 T_DOUBLEFLT */ 157 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 158 "invalid TSS fault", /* 25 T_TSSFLT */ 159 "segment not present fault", /* 26 T_SEGNPFLT */ 160 "stack fault", /* 27 T_STKFLT */ 161 "machine check trap", /* 28 T_MCHK */ 162 "SIMD floating-point exception", /* 29 T_XMMFLT */ 163 "reserved (unknown) fault", /* 30 T_RESERVED */ 164 "", /* 31 unused (reserved) */ 165 "DTrace pid return trap", /* 32 T_DTRACE_RET */ 166 "DTrace fasttrap probe trap", /* 33 T_DTRACE_PROBE */ 167}; 168 169#ifdef KDB 170static int kdb_on_nmi = 1; 171SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW, 172 &kdb_on_nmi, 0, "Go to KDB on NMI"); 173TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi); 174#endif 175static int panic_on_nmi = 1; 176SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 177 &panic_on_nmi, 0, "Panic on NMI"); 178TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi); 179static int prot_fault_translation; 180SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW, 181 &prot_fault_translation, 0, 182 "Select signal to deliver on protection fault"); 183static int uprintf_signal; 184SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW, 185 &uprintf_signal, 0, 186 "Print debugging information on trap signal to ctty"); 187 188/* 189 * Exception, fault, and trap interface to the FreeBSD kernel. 190 * This common code is called from assembly language IDT gate entry 191 * routines that prepare a suitable stack frame, and restore this 192 * frame after the exception has been processed. 193 */ 194 195void 196trap(struct trapframe *frame) 197{ 198 struct thread *td = curthread; 199 struct proc *p = td->td_proc; 200 int i = 0, ucode = 0, code; 201 u_int type; 202 register_t addr = 0; 203 ksiginfo_t ksi; 204 205 PCPU_INC(cnt.v_trap); 206 type = frame->tf_trapno; 207 208#ifdef SMP 209 /* Handler for NMI IPIs used for stopping CPUs. */ 210 if (type == T_NMI) { 211 if (ipi_nmi_handler() == 0) 212 goto out; 213 } 214#endif /* SMP */ 215 216#ifdef KDB 217 if (kdb_active) { 218 kdb_reenter(); 219 goto out; 220 } 221#endif 222 223 if (type == T_RESERVED) { 224 trap_fatal(frame, 0); 225 goto out; 226 } 227 228#ifdef HWPMC_HOOKS 229 /* 230 * CPU PMCs interrupt using an NMI. If the PMC module is 231 * active, pass the 'rip' value to the PMC module's interrupt 232 * handler. A return value of '1' from the handler means that 233 * the NMI was handled by it and we can return immediately. 234 */ 235 if (type == T_NMI && pmc_intr && 236 (*pmc_intr)(PCPU_GET(cpuid), frame)) 237 goto out; 238#endif 239 240 if (type == T_MCHK) { 241 mca_intr(); 242 goto out; 243 } 244 245#ifdef KDTRACE_HOOKS 246 /* 247 * A trap can occur while DTrace executes a probe. Before 248 * executing the probe, DTrace blocks re-scheduling and sets 249 * a flag in it's per-cpu flags to indicate that it doesn't 250 * want to fault. On returning from the probe, the no-fault 251 * flag is cleared and finally re-scheduling is enabled. 252 * 253 * If the DTrace kernel module has registered a trap handler, 254 * call it and if it returns non-zero, assume that it has 255 * handled the trap and modified the trap frame so that this 256 * function can return normally. 257 */ 258 if (type == T_DTRACE_PROBE || type == T_DTRACE_RET || 259 type == T_BPTFLT) { 260 struct reg regs; 261 262 fill_frame_regs(frame, ®s); 263 if (type == T_DTRACE_PROBE && 264 dtrace_fasttrap_probe_ptr != NULL && 265 dtrace_fasttrap_probe_ptr(®s) == 0) 266 goto out; 267 else if (type == T_BPTFLT && 268 dtrace_pid_probe_ptr != NULL && 269 dtrace_pid_probe_ptr(®s) == 0) 270 goto out; 271 else if (type == T_DTRACE_RET && 272 dtrace_return_probe_ptr != NULL && 273 dtrace_return_probe_ptr(®s) == 0) 274 goto out; 275 } 276 if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type)) 277 goto out; 278#endif 279 280 if ((frame->tf_rflags & PSL_I) == 0) { 281 /* 282 * Buggy application or kernel code has disabled 283 * interrupts and then trapped. Enabling interrupts 284 * now is wrong, but it is better than running with 285 * interrupts disabled until they are accidentally 286 * enabled later. 287 */ 288 if (ISPL(frame->tf_cs) == SEL_UPL) 289 uprintf( 290 "pid %ld (%s): trap %d with interrupts disabled\n", 291 (long)curproc->p_pid, curthread->td_name, type); 292 else if (type != T_NMI && type != T_BPTFLT && 293 type != T_TRCTRAP) { 294 /* 295 * XXX not quite right, since this may be for a 296 * multiple fault in user mode. 297 */ 298 printf("kernel trap %d with interrupts disabled\n", 299 type); 300 301 /* 302 * We shouldn't enable interrupts while holding a 303 * spin lock. 304 */ 305 if (td->td_md.md_spinlock_count == 0) 306 enable_intr(); 307 } 308 } 309 310 code = frame->tf_err; 311 312 if (ISPL(frame->tf_cs) == SEL_UPL) { 313 /* user trap */ 314 315 td->td_pticks = 0; 316 td->td_frame = frame; 317 addr = frame->tf_rip; 318 if (td->td_ucred != p->p_ucred) 319 cred_update_thread(td); 320 321 switch (type) { 322 case T_PRIVINFLT: /* privileged instruction fault */ 323 i = SIGILL; 324 ucode = ILL_PRVOPC; 325 break; 326 327 case T_BPTFLT: /* bpt instruction fault */ 328 case T_TRCTRAP: /* trace trap */ 329 enable_intr(); 330 frame->tf_rflags &= ~PSL_T; 331 i = SIGTRAP; 332 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 333 break; 334 335 case T_ARITHTRAP: /* arithmetic trap */ 336 ucode = fputrap_x87(); 337 if (ucode == -1) 338 goto userout; 339 i = SIGFPE; 340 break; 341 342 case T_PROTFLT: /* general protection fault */ 343 i = SIGBUS; 344 ucode = BUS_OBJERR; 345 break; 346 case T_STKFLT: /* stack fault */ 347 case T_SEGNPFLT: /* segment not present fault */ 348 i = SIGBUS; 349 ucode = BUS_ADRERR; 350 break; 351 case T_TSSFLT: /* invalid TSS fault */ 352 i = SIGBUS; 353 ucode = BUS_OBJERR; 354 break; 355 case T_DOUBLEFLT: /* double fault */ 356 default: 357 i = SIGBUS; 358 ucode = BUS_OBJERR; 359 break; 360 361 case T_PAGEFLT: /* page fault */ 362 addr = frame->tf_addr; 363 i = trap_pfault(frame, TRUE); 364 if (i == -1) 365 goto userout; 366 if (i == 0) 367 goto user; 368 369 if (i == SIGSEGV) 370 ucode = SEGV_MAPERR; 371 else { 372 if (prot_fault_translation == 0) { 373 /* 374 * Autodetect. 375 * This check also covers the images 376 * without the ABI-tag ELF note. 377 */ 378 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD 379 && p->p_osrel >= P_OSREL_SIGSEGV) { 380 i = SIGSEGV; 381 ucode = SEGV_ACCERR; 382 } else { 383 i = SIGBUS; 384 ucode = BUS_PAGE_FAULT; 385 } 386 } else if (prot_fault_translation == 1) { 387 /* 388 * Always compat mode. 389 */ 390 i = SIGBUS; 391 ucode = BUS_PAGE_FAULT; 392 } else { 393 /* 394 * Always SIGSEGV mode. 395 */ 396 i = SIGSEGV; 397 ucode = SEGV_ACCERR; 398 } 399 } 400 break; 401 402 case T_DIVIDE: /* integer divide fault */ 403 ucode = FPE_INTDIV; 404 i = SIGFPE; 405 break; 406 407#ifdef DEV_ISA 408 case T_NMI: 409 /* machine/parity/power fail/"kitchen sink" faults */ 410 if (isa_nmi(code) == 0) { 411#ifdef KDB 412 /* 413 * NMI can be hooked up to a pushbutton 414 * for debugging. 415 */ 416 if (kdb_on_nmi) { 417 printf ("NMI ... going to debugger\n"); 418 kdb_trap(type, 0, frame); 419 } 420#endif /* KDB */ 421 goto userout; 422 } else if (panic_on_nmi) 423 panic("NMI indicates hardware failure"); 424 break; 425#endif /* DEV_ISA */ 426 427 case T_OFLOW: /* integer overflow fault */ 428 ucode = FPE_INTOVF; 429 i = SIGFPE; 430 break; 431 432 case T_BOUND: /* bounds check fault */ 433 ucode = FPE_FLTSUB; 434 i = SIGFPE; 435 break; 436 437 case T_DNA: 438 /* transparent fault (due to context switch "late") */ 439 KASSERT(PCB_USER_FPU(td->td_pcb), 440 ("kernel FPU ctx has leaked")); 441 fpudna(); 442 goto userout; 443 444 case T_FPOPFLT: /* FPU operand fetch fault */ 445 ucode = ILL_COPROC; 446 i = SIGILL; 447 break; 448 449 case T_XMMFLT: /* SIMD floating-point exception */ 450 ucode = fputrap_sse(); 451 if (ucode == -1) 452 goto userout; 453 i = SIGFPE; 454 break; 455 } 456 } else { 457 /* kernel trap */ 458 459 KASSERT(cold || td->td_ucred != NULL, 460 ("kernel trap doesn't have ucred")); 461 switch (type) { 462 case T_PAGEFLT: /* page fault */ 463 (void) trap_pfault(frame, FALSE); 464 goto out; 465 466 case T_DNA: 467 KASSERT(!PCB_USER_FPU(td->td_pcb), 468 ("Unregistered use of FPU in kernel")); 469 fpudna(); 470 goto out; 471 472 case T_ARITHTRAP: /* arithmetic trap */ 473 case T_XMMFLT: /* SIMD floating-point exception */ 474 case T_FPOPFLT: /* FPU operand fetch fault */ 475 /* 476 * XXXKIB for now disable any FPU traps in kernel 477 * handler registration seems to be overkill 478 */ 479 trap_fatal(frame, 0); 480 goto out; 481 482 case T_STKFLT: /* stack fault */ 483 break; 484 485 case T_PROTFLT: /* general protection fault */ 486 case T_SEGNPFLT: /* segment not present fault */ 487 if (td->td_intr_nesting_level != 0) 488 break; 489 490 /* 491 * Invalid segment selectors and out of bounds 492 * %rip's and %rsp's can be set up in user mode. 493 * This causes a fault in kernel mode when the 494 * kernel tries to return to user mode. We want 495 * to get this fault so that we can fix the 496 * problem here and not have to check all the 497 * selectors and pointers when the user changes 498 * them. 499 */ 500 if (frame->tf_rip == (long)doreti_iret) { 501 frame->tf_rip = (long)doreti_iret_fault; 502 goto out; 503 } 504 if (frame->tf_rip == (long)ld_ds) { 505 frame->tf_rip = (long)ds_load_fault; 506 goto out; 507 } 508 if (frame->tf_rip == (long)ld_es) { 509 frame->tf_rip = (long)es_load_fault; 510 goto out; 511 } 512 if (frame->tf_rip == (long)ld_fs) { 513 frame->tf_rip = (long)fs_load_fault; 514 goto out; 515 } 516 if (frame->tf_rip == (long)ld_gs) { 517 frame->tf_rip = (long)gs_load_fault; 518 goto out; 519 } 520 if (frame->tf_rip == (long)ld_gsbase) { 521 frame->tf_rip = (long)gsbase_load_fault; 522 goto out; 523 } 524 if (frame->tf_rip == (long)ld_fsbase) { 525 frame->tf_rip = (long)fsbase_load_fault; 526 goto out; 527 } 528 if (curpcb->pcb_onfault != NULL) { 529 frame->tf_rip = (long)curpcb->pcb_onfault; 530 goto out; 531 } 532 break; 533 534 case T_TSSFLT: 535 /* 536 * PSL_NT can be set in user mode and isn't cleared 537 * automatically when the kernel is entered. This 538 * causes a TSS fault when the kernel attempts to 539 * `iret' because the TSS link is uninitialized. We 540 * want to get this fault so that we can fix the 541 * problem here and not every time the kernel is 542 * entered. 543 */ 544 if (frame->tf_rflags & PSL_NT) { 545 frame->tf_rflags &= ~PSL_NT; 546 goto out; 547 } 548 break; 549 550 case T_TRCTRAP: /* trace trap */ 551 /* 552 * Ignore debug register trace traps due to 553 * accesses in the user's address space, which 554 * can happen under several conditions such as 555 * if a user sets a watchpoint on a buffer and 556 * then passes that buffer to a system call. 557 * We still want to get TRCTRAPS for addresses 558 * in kernel space because that is useful when 559 * debugging the kernel. 560 */ 561 if (user_dbreg_trap()) { 562 /* 563 * Reset breakpoint bits because the 564 * processor doesn't 565 */ 566 /* XXX check upper bits here */ 567 load_dr6(rdr6() & 0xfffffff0); 568 goto out; 569 } 570 /* 571 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 572 */ 573 case T_BPTFLT: 574 /* 575 * If KDB is enabled, let it handle the debugger trap. 576 * Otherwise, debugger traps "can't happen". 577 */ 578#ifdef KDB 579 if (kdb_trap(type, 0, frame)) 580 goto out; 581#endif 582 break; 583 584#ifdef DEV_ISA 585 case T_NMI: 586 /* machine/parity/power fail/"kitchen sink" faults */ 587 if (isa_nmi(code) == 0) { 588#ifdef KDB 589 /* 590 * NMI can be hooked up to a pushbutton 591 * for debugging. 592 */ 593 if (kdb_on_nmi) { 594 printf ("NMI ... going to debugger\n"); 595 kdb_trap(type, 0, frame); 596 } 597#endif /* KDB */ 598 goto out; 599 } else if (panic_on_nmi == 0) 600 goto out; 601 /* FALLTHROUGH */ 602#endif /* DEV_ISA */ 603 } 604 605 trap_fatal(frame, 0); 606 goto out; 607 } 608 609 /* Translate fault for emulators (e.g. Linux) */ 610 if (*p->p_sysent->sv_transtrap) 611 i = (*p->p_sysent->sv_transtrap)(i, type); 612 613 ksiginfo_init_trap(&ksi); 614 ksi.ksi_signo = i; 615 ksi.ksi_code = ucode; 616 ksi.ksi_trapno = type; 617 ksi.ksi_addr = (void *)addr; 618 if (uprintf_signal) { 619 uprintf("pid %d comm %s: signal %d err %lx code %d type %d " 620 "addr 0x%lx rsp 0x%lx rip 0x%lx " 621 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n", 622 p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr, 623 frame->tf_rsp, frame->tf_rip, 624 fubyte((void *)(frame->tf_rip + 0)), 625 fubyte((void *)(frame->tf_rip + 1)), 626 fubyte((void *)(frame->tf_rip + 2)), 627 fubyte((void *)(frame->tf_rip + 3)), 628 fubyte((void *)(frame->tf_rip + 4)), 629 fubyte((void *)(frame->tf_rip + 5)), 630 fubyte((void *)(frame->tf_rip + 6)), 631 fubyte((void *)(frame->tf_rip + 7))); 632 } 633 KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled")); 634 trapsignal(td, &ksi); 635 636user: 637 userret(td, frame); 638 KASSERT(PCB_USER_FPU(td->td_pcb), 639 ("Return from trap with kernel FPU ctx leaked")); 640userout: 641out: 642 return; 643} 644 645static int 646trap_pfault(frame, usermode) 647 struct trapframe *frame; 648 int usermode; 649{ 650 vm_offset_t va; 651 struct vmspace *vm; 652 vm_map_t map; 653 int rv = 0; 654 vm_prot_t ftype; 655 struct thread *td = curthread; 656 struct proc *p = td->td_proc; 657 vm_offset_t eva = frame->tf_addr; 658 659 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) { 660 /* 661 * Due to both processor errata and lazy TLB invalidation when 662 * access restrictions are removed from virtual pages, memory 663 * accesses that are allowed by the physical mapping layer may 664 * nonetheless cause one spurious page fault per virtual page. 665 * When the thread is executing a "no faulting" section that 666 * is bracketed by vm_fault_{disable,enable}_pagefaults(), 667 * every page fault is treated as a spurious page fault, 668 * unless it accesses the same virtual address as the most 669 * recent page fault within the same "no faulting" section. 670 */ 671 if (td->td_md.md_spurflt_addr != eva || 672 (td->td_pflags & TDP_RESETSPUR) != 0) { 673 /* 674 * Do nothing to the TLB. A stale TLB entry is 675 * flushed automatically by a page fault. 676 */ 677 td->td_md.md_spurflt_addr = eva; 678 td->td_pflags &= ~TDP_RESETSPUR; 679 return (0); 680 } 681 } else { 682 /* 683 * If we get a page fault while in a critical section, then 684 * it is most likely a fatal kernel page fault. The kernel 685 * is already going to panic trying to get a sleep lock to 686 * do the VM lookup, so just consider it a fatal trap so the 687 * kernel can print out a useful trap message and even get 688 * to the debugger. 689 * 690 * If we get a page fault while holding a non-sleepable 691 * lock, then it is most likely a fatal kernel page fault. 692 * If WITNESS is enabled, then it's going to whine about 693 * bogus LORs with various VM locks, so just skip to the 694 * fatal trap handling directly. 695 */ 696 if (td->td_critnest != 0 || 697 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL, 698 "Kernel page fault") != 0) { 699 trap_fatal(frame, eva); 700 return (-1); 701 } 702 } 703 va = trunc_page(eva); 704 if (va >= VM_MIN_KERNEL_ADDRESS) { 705 /* 706 * Don't allow user-mode faults in kernel address space. 707 */ 708 if (usermode) 709 goto nogo; 710 711 map = kernel_map; 712 } else { 713 /* 714 * This is a fault on non-kernel virtual memory. If either 715 * p or p->p_vmspace is NULL, then the fault is fatal. 716 */ 717 if (p == NULL || (vm = p->p_vmspace) == NULL) 718 goto nogo; 719 720 map = &vm->vm_map; 721 722 /* 723 * When accessing a usermode address, kernel must be 724 * ready to accept the page fault, and provide a 725 * handling routine. Since accessing the address 726 * without the handler is a bug, do not try to handle 727 * it normally, and panic immediately. 728 */ 729 if (!usermode && (td->td_intr_nesting_level != 0 || 730 curpcb->pcb_onfault == NULL)) { 731 trap_fatal(frame, eva); 732 return (-1); 733 } 734 } 735 736 /* 737 * If the trap was caused by errant bits in the PTE then panic. 738 */ 739 if (frame->tf_err & PGEX_RSV) { 740 trap_fatal(frame, eva); 741 return (-1); 742 } 743 744 /* 745 * PGEX_I is defined only if the execute disable bit capability is 746 * supported and enabled. 747 */ 748 if (frame->tf_err & PGEX_W) 749 ftype = VM_PROT_WRITE; 750 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 751 ftype = VM_PROT_EXECUTE; 752 else 753 ftype = VM_PROT_READ; 754 755 if (map != kernel_map) { 756 /* 757 * Keep swapout from messing with us during this 758 * critical time. 759 */ 760 PROC_LOCK(p); 761 ++p->p_lock; 762 PROC_UNLOCK(p); 763 764 /* Fault in the user page: */ 765 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 766 767 PROC_LOCK(p); 768 --p->p_lock; 769 PROC_UNLOCK(p); 770 } else { 771 /* 772 * Don't have to worry about process locking or stacks in the 773 * kernel. 774 */ 775 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 776 } 777 if (rv == KERN_SUCCESS) { 778#ifdef HWPMC_HOOKS 779 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 780 PMC_SOFT_CALL_TF( , , page_fault, all, frame); 781 if (ftype == VM_PROT_READ) 782 PMC_SOFT_CALL_TF( , , page_fault, read, 783 frame); 784 else 785 PMC_SOFT_CALL_TF( , , page_fault, write, 786 frame); 787 } 788#endif 789 return (0); 790 } 791nogo: 792 if (!usermode) { 793 if (td->td_intr_nesting_level == 0 && 794 curpcb->pcb_onfault != NULL) { 795 frame->tf_rip = (long)curpcb->pcb_onfault; 796 return (0); 797 } 798 trap_fatal(frame, eva); 799 return (-1); 800 } 801 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 802} 803 804static void 805trap_fatal(frame, eva) 806 struct trapframe *frame; 807 vm_offset_t eva; 808{ 809 int code, ss; 810 u_int type; 811 long esp; 812 struct soft_segment_descriptor softseg; 813 char *msg; 814 815 code = frame->tf_err; 816 type = frame->tf_trapno; 817 sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)], 818 &softseg); 819 820 if (type <= MAX_TRAP_MSG) 821 msg = trap_msg[type]; 822 else 823 msg = "UNKNOWN"; 824 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 825 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 826#ifdef SMP 827 /* two separate prints in case of a trap on an unmapped page */ 828 printf("cpuid = %d; ", PCPU_GET(cpuid)); 829 printf("apic id = %02x\n", PCPU_GET(apic_id)); 830#endif 831 if (type == T_PAGEFLT) { 832 printf("fault virtual address = 0x%lx\n", eva); 833 printf("fault code = %s %s %s%s, %s\n", 834 code & PGEX_U ? "user" : "supervisor", 835 code & PGEX_W ? "write" : "read", 836 code & PGEX_I ? "instruction" : "data", 837 code & PGEX_RSV ? " rsv" : "", 838 code & PGEX_P ? "protection violation" : "page not present"); 839 } 840 printf("instruction pointer = 0x%lx:0x%lx\n", 841 frame->tf_cs & 0xffff, frame->tf_rip); 842 if (ISPL(frame->tf_cs) == SEL_UPL) { 843 ss = frame->tf_ss & 0xffff; 844 esp = frame->tf_rsp; 845 } else { 846 ss = GSEL(GDATA_SEL, SEL_KPL); 847 esp = (long)&frame->tf_rsp; 848 } 849 printf("stack pointer = 0x%x:0x%lx\n", ss, esp); 850 printf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 851 printf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 852 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 853 printf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 854 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 855 softseg.ssd_gran); 856 printf("processor eflags = "); 857 if (frame->tf_rflags & PSL_T) 858 printf("trace trap, "); 859 if (frame->tf_rflags & PSL_I) 860 printf("interrupt enabled, "); 861 if (frame->tf_rflags & PSL_NT) 862 printf("nested task, "); 863 if (frame->tf_rflags & PSL_RF) 864 printf("resume, "); 865 printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 866 printf("current process = "); 867 if (curproc) { 868 printf("%lu (%s)\n", 869 (u_long)curproc->p_pid, curthread->td_name ? 870 curthread->td_name : ""); 871 } else { 872 printf("Idle\n"); 873 } 874 875#ifdef KDB 876 if (debugger_on_panic || kdb_active) 877 if (kdb_trap(type, 0, frame)) 878 return; 879#endif 880 printf("trap number = %d\n", type); 881 if (type <= MAX_TRAP_MSG) 882 panic("%s", trap_msg[type]); 883 else 884 panic("unknown/reserved trap"); 885} 886 887/* 888 * Double fault handler. Called when a fault occurs while writing 889 * a frame for a trap/exception onto the stack. This usually occurs 890 * when the stack overflows (such is the case with infinite recursion, 891 * for example). 892 */ 893void 894dblfault_handler(struct trapframe *frame) 895{ 896#ifdef KDTRACE_HOOKS 897 if (dtrace_doubletrap_func != NULL) 898 (*dtrace_doubletrap_func)(); 899#endif 900 printf("\nFatal double fault\n"); 901 printf("rip = 0x%lx\n", frame->tf_rip); 902 printf("rsp = 0x%lx\n", frame->tf_rsp); 903 printf("rbp = 0x%lx\n", frame->tf_rbp); 904#ifdef SMP 905 /* two separate prints in case of a trap on an unmapped page */ 906 printf("cpuid = %d; ", PCPU_GET(cpuid)); 907 printf("apic id = %02x\n", PCPU_GET(apic_id)); 908#endif 909 panic("double fault"); 910} 911 912int 913cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa) 914{ 915 struct proc *p; 916 struct trapframe *frame; 917 register_t *argp; 918 caddr_t params; 919 int reg, regcnt, error; 920 921 p = td->td_proc; 922 frame = td->td_frame; 923 reg = 0; 924 regcnt = 6; 925 926 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 927 sa->code = frame->tf_rax; 928 929 if (sa->code == SYS_syscall || sa->code == SYS___syscall) { 930 sa->code = frame->tf_rdi; 931 reg++; 932 regcnt--; 933 } 934 if (p->p_sysent->sv_mask) 935 sa->code &= p->p_sysent->sv_mask; 936 937 if (sa->code >= p->p_sysent->sv_size) 938 sa->callp = &p->p_sysent->sv_table[0]; 939 else 940 sa->callp = &p->p_sysent->sv_table[sa->code]; 941 942 sa->narg = sa->callp->sy_narg; 943 KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]), 944 ("Too many syscall arguments!")); 945 error = 0; 946 argp = &frame->tf_rdi; 947 argp += reg; 948 bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt); 949 if (sa->narg > regcnt) { 950 KASSERT(params != NULL, ("copyin args with no params!")); 951 error = copyin(params, &sa->args[regcnt], 952 (sa->narg - regcnt) * sizeof(sa->args[0])); 953 } 954 955 if (error == 0) { 956 td->td_retval[0] = 0; 957 td->td_retval[1] = frame->tf_rdx; 958 } 959 960 return (error); 961} 962 963#include "../../kern/subr_syscall.c" 964 965/* 966 * System call handler for native binaries. The trap frame is already 967 * set up by the assembler trampoline and a pointer to it is saved in 968 * td_frame. 969 */ 970void 971amd64_syscall(struct thread *td, int traced) 972{ 973 struct syscall_args sa; 974 int error; 975 ksiginfo_t ksi; 976 977#ifdef DIAGNOSTIC 978 if (ISPL(td->td_frame->tf_cs) != SEL_UPL) { 979 panic("syscall"); 980 /* NOT REACHED */ 981 } 982#endif 983 error = syscallenter(td, &sa); 984 985 /* 986 * Traced syscall. 987 */ 988 if (__predict_false(traced)) { 989 td->td_frame->tf_rflags &= ~PSL_T; 990 ksiginfo_init_trap(&ksi); 991 ksi.ksi_signo = SIGTRAP; 992 ksi.ksi_code = TRAP_TRACE; 993 ksi.ksi_addr = (void *)td->td_frame->tf_rip; 994 trapsignal(td, &ksi); 995 } 996 997 KASSERT(PCB_USER_FPU(td->td_pcb), 998 ("System call %s returing with kernel FPU ctx leaked", 999 syscallname(td->td_proc, sa.code))); 1000 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td), 1001 ("System call %s returning with mangled pcb_save", 1002 syscallname(td->td_proc, sa.code))); 1003 1004 syscallret(td, error, &sa); 1005 1006 /* 1007 * If the user-supplied value of %rip is not a canonical 1008 * address, then some CPUs will trigger a ring 0 #GP during 1009 * the sysret instruction. However, the fault handler would 1010 * execute in ring 0 with the user's %gs and %rsp which would 1011 * not be safe. Instead, use the full return path which 1012 * catches the problem safely. 1013 */ 1014 if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS) 1015 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 1016} 1017