1 /*- 2 * Copyright (c) 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (C) 1994, David Greenman 5 * Copyright (c) 2008 The DragonFly Project. 6 * Copyright (c) 2008 Jordan Gordeev. 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 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $ 41 */ 42 43 /* 44 * x86_64 Trap and System call handling 45 */ 46 47 #include "opt_ddb.h" 48 #include "opt_ktrace.h" 49 50 #include <machine/frame.h> 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/kernel.h> 54 #include <sys/kerneldump.h> 55 #include <sys/proc.h> 56 #include <sys/pioctl.h> 57 #include <sys/types.h> 58 #include <sys/signal2.h> 59 #include <sys/syscall.h> 60 #include <sys/sysctl.h> 61 #include <sys/sysent.h> 62 #include <sys/systm.h> 63 #ifdef KTRACE 64 #include <sys/ktrace.h> 65 #endif 66 #include <sys/ktr.h> 67 #include <sys/sysmsg.h> 68 #include <sys/sysproto.h> 69 #include <sys/sysunion.h> 70 71 #include <vm/pmap.h> 72 #include <vm/vm.h> 73 #include <vm/vm_extern.h> 74 #include <vm/vm_kern.h> 75 #include <vm/vm_param.h> 76 #include <machine/cpu.h> 77 #include <machine/pcb.h> 78 #include <machine/smp.h> 79 #include <machine/thread.h> 80 #include <machine/vmparam.h> 81 #include <machine/md_var.h> 82 #include <machine_base/isa/intr_machdep.h> 83 84 #include <ddb/ddb.h> 85 86 #include <sys/thread2.h> 87 #include <sys/mplock2.h> 88 89 #ifdef SMP 90 91 #define MAKEMPSAFE(have_mplock) \ 92 if (have_mplock == 0) { \ 93 get_mplock(); \ 94 have_mplock = 1; \ 95 } 96 97 #else 98 99 #define MAKEMPSAFE(have_mplock) 100 101 #endif 102 103 extern void trap(struct trapframe *frame); 104 105 static int trap_pfault(struct trapframe *, int); 106 static void trap_fatal(struct trapframe *, vm_offset_t); 107 void dblfault_handler(struct trapframe *frame); 108 109 #define MAX_TRAP_MSG 30 110 static char *trap_msg[] = { 111 "", /* 0 unused */ 112 "privileged instruction fault", /* 1 T_PRIVINFLT */ 113 "", /* 2 unused */ 114 "breakpoint instruction fault", /* 3 T_BPTFLT */ 115 "", /* 4 unused */ 116 "", /* 5 unused */ 117 "arithmetic trap", /* 6 T_ARITHTRAP */ 118 "system forced exception", /* 7 T_ASTFLT */ 119 "", /* 8 unused */ 120 "general protection fault", /* 9 T_PROTFLT */ 121 "trace trap", /* 10 T_TRCTRAP */ 122 "", /* 11 unused */ 123 "page fault", /* 12 T_PAGEFLT */ 124 "", /* 13 unused */ 125 "alignment fault", /* 14 T_ALIGNFLT */ 126 "", /* 15 unused */ 127 "", /* 16 unused */ 128 "", /* 17 unused */ 129 "integer divide fault", /* 18 T_DIVIDE */ 130 "non-maskable interrupt trap", /* 19 T_NMI */ 131 "overflow trap", /* 20 T_OFLOW */ 132 "FPU bounds check fault", /* 21 T_BOUND */ 133 "FPU device not available", /* 22 T_DNA */ 134 "double fault", /* 23 T_DOUBLEFLT */ 135 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 136 "invalid TSS fault", /* 25 T_TSSFLT */ 137 "segment not present fault", /* 26 T_SEGNPFLT */ 138 "stack fault", /* 27 T_STKFLT */ 139 "machine check trap", /* 28 T_MCHK */ 140 "SIMD floating-point exception", /* 29 T_XMMFLT */ 141 "reserved (unknown) fault", /* 30 T_RESERVED */ 142 }; 143 144 #ifdef DDB 145 static int ddb_on_nmi = 1; 146 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 147 &ddb_on_nmi, 0, "Go to DDB on NMI"); 148 static int ddb_on_seg_fault = 0; 149 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW, 150 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault"); 151 #endif 152 static int panic_on_nmi = 1; 153 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 154 &panic_on_nmi, 0, "Panic on NMI"); 155 static int fast_release; 156 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 157 &fast_release, 0, "Passive Release was optimal"); 158 static int slow_release; 159 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 160 &slow_release, 0, "Passive Release was nonoptimal"); 161 162 /* 163 * Passively intercepts the thread switch function to increase 164 * the thread priority from a user priority to a kernel priority, reducing 165 * syscall and trap overhead for the case where no switch occurs. 166 * 167 * Synchronizes td_ucred with p_ucred. This is used by system calls, 168 * signal handling, faults, AST traps, and anything else that enters the 169 * kernel from userland and provides the kernel with a stable read-only 170 * copy of the process ucred. 171 */ 172 static __inline void 173 userenter(struct thread *curtd, struct proc *curp) 174 { 175 struct ucred *ocred; 176 struct ucred *ncred; 177 178 curtd->td_release = lwkt_passive_release; 179 180 if (curtd->td_ucred != curp->p_ucred) { 181 ncred = crhold(curp->p_ucred); 182 ocred = curtd->td_ucred; 183 curtd->td_ucred = ncred; 184 if (ocred) 185 crfree(ocred); 186 } 187 } 188 189 /* 190 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 191 * must be completed before we can return to or try to return to userland. 192 * 193 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 194 * arithmatic on the delta calculation so the absolute tick values are 195 * truncated to an integer. 196 */ 197 static void 198 userret(struct lwp *lp, struct trapframe *frame, int sticks) 199 { 200 struct proc *p = lp->lwp_proc; 201 int sig; 202 203 /* 204 * Charge system time if profiling. Note: times are in microseconds. 205 * This may do a copyout and block, so do it first even though it 206 * means some system time will be charged as user time. 207 */ 208 if (p->p_flag & P_PROFIL) { 209 addupc_task(p, frame->tf_rip, 210 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 211 } 212 213 recheck: 214 /* 215 * If the jungle wants us dead, so be it. 216 */ 217 if (lp->lwp_flag & LWP_WEXIT) { 218 get_mplock(); 219 lwp_exit(0); 220 rel_mplock(); /* NOT REACHED */ 221 } 222 223 /* 224 * Block here if we are in a stopped state. 225 */ 226 if (p->p_stat == SSTOP || dump_stop_usertds) { 227 get_mplock(); 228 tstop(); 229 rel_mplock(); 230 goto recheck; 231 } 232 233 /* 234 * Post any pending upcalls. If running a virtual kernel be sure 235 * to restore the virtual kernel's vmspace before posting the upcall. 236 */ 237 if (p->p_flag & P_UPCALLPEND) { 238 p->p_flag &= ~P_UPCALLPEND; 239 get_mplock(); 240 postupcall(lp); 241 rel_mplock(); 242 goto recheck; 243 } 244 245 /* 246 * Post any pending signals. If running a virtual kernel be sure 247 * to restore the virtual kernel's vmspace before posting the signal. 248 * 249 * WARNING! postsig() can exit and not return. 250 */ 251 if ((sig = CURSIG_TRACE(lp)) != 0) { 252 get_mplock(); 253 postsig(sig); 254 rel_mplock(); 255 goto recheck; 256 } 257 258 /* 259 * block here if we are swapped out, but still process signals 260 * (such as SIGKILL). proc0 (the swapin scheduler) is already 261 * aware of our situation, we do not have to wake it up. 262 */ 263 if (p->p_flag & P_SWAPPEDOUT) { 264 get_mplock(); 265 p->p_flag |= P_SWAPWAIT; 266 swapin_request(); 267 if (p->p_flag & P_SWAPWAIT) 268 tsleep(p, PCATCH, "SWOUT", 0); 269 p->p_flag &= ~P_SWAPWAIT; 270 rel_mplock(); 271 goto recheck; 272 } 273 274 /* 275 * Make sure postsig() handled request to restore old signal mask after 276 * running signal handler. 277 */ 278 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0); 279 } 280 281 /* 282 * Cleanup from userenter and any passive release that might have occured. 283 * We must reclaim the current-process designation before we can return 284 * to usermode. We also handle both LWKT and USER reschedule requests. 285 */ 286 static __inline void 287 userexit(struct lwp *lp) 288 { 289 struct thread *td = lp->lwp_thread; 290 /* globaldata_t gd = td->td_gd;*/ 291 292 /* 293 * Handle stop requests at kernel priority. Any requests queued 294 * after this loop will generate another AST. 295 */ 296 while (lp->lwp_proc->p_stat == SSTOP) { 297 get_mplock(); 298 tstop(); 299 rel_mplock(); 300 } 301 302 /* 303 * Reduce our priority in preparation for a return to userland. If 304 * our passive release function was still in place, our priority was 305 * never raised and does not need to be reduced. 306 */ 307 lwkt_passive_recover(td); 308 309 /* 310 * Become the current user scheduled process if we aren't already, 311 * and deal with reschedule requests and other factors. 312 */ 313 lp->lwp_proc->p_usched->acquire_curproc(lp); 314 /* WARNING: we may have migrated cpu's */ 315 /* gd = td->td_gd; */ 316 } 317 318 #if !defined(KTR_KERNENTRY) 319 #define KTR_KERNENTRY KTR_ALL 320 #endif 321 KTR_INFO_MASTER(kernentry); 322 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR", 323 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t)); 324 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR", 325 sizeof(long) + sizeof(long)); 326 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR", 327 sizeof(long) + sizeof(long) + sizeof(long)); 328 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR", 329 sizeof(long) + sizeof(long) + sizeof(long)); 330 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR", 331 sizeof(long) + sizeof(long)); 332 333 /* 334 * Exception, fault, and trap interface to the kernel. 335 * This common code is called from assembly language IDT gate entry 336 * routines that prepare a suitable stack frame, and restore this 337 * frame after the exception has been processed. 338 * 339 * This function is also called from doreti in an interlock to handle ASTs. 340 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 341 * 342 * NOTE! We have to retrieve the fault address prior to obtaining the 343 * MP lock because get_mplock() may switch out. YYY cr2 really ought 344 * to be retrieved by the assembly code, not here. 345 * 346 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 347 * if an attempt is made to switch from a fast interrupt or IPI. This is 348 * necessary to properly take fatal kernel traps on SMP machines if 349 * get_mplock() has to block. 350 */ 351 352 void 353 trap(struct trapframe *frame) 354 { 355 struct globaldata *gd = mycpu; 356 struct thread *td = gd->gd_curthread; 357 struct lwp *lp = td->td_lwp; 358 struct proc *p; 359 int sticks = 0; 360 int i = 0, ucode = 0, type, code; 361 #ifdef SMP 362 int have_mplock = 0; 363 #endif 364 #ifdef INVARIANTS 365 int crit_count = td->td_critcount; 366 lwkt_tokref_t curstop = td->td_toks_stop; 367 #endif 368 vm_offset_t eva; 369 370 p = td->td_proc; 371 372 #ifdef DDB 373 /* 374 * We need to allow T_DNA faults when the debugger is active since 375 * some dumping paths do large bcopy() which use the floating 376 * point registers for faster copying. 377 */ 378 if (db_active && frame->tf_trapno != T_DNA) { 379 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 380 ++gd->gd_trap_nesting_level; 381 MAKEMPSAFE(have_mplock); 382 trap_fatal(frame, eva); 383 --gd->gd_trap_nesting_level; 384 goto out2; 385 } 386 #endif 387 388 eva = 0; 389 390 if ((frame->tf_rflags & PSL_I) == 0) { 391 /* 392 * Buggy application or kernel code has disabled interrupts 393 * and then trapped. Enabling interrupts now is wrong, but 394 * it is better than running with interrupts disabled until 395 * they are accidentally enabled later. 396 */ 397 type = frame->tf_trapno; 398 if (ISPL(frame->tf_cs) == SEL_UPL) { 399 MAKEMPSAFE(have_mplock); 400 /* JG curproc can be NULL */ 401 kprintf( 402 "pid %ld (%s): trap %d with interrupts disabled\n", 403 (long)curproc->p_pid, curproc->p_comm, type); 404 } else if (type != T_NMI && type != T_BPTFLT && 405 type != T_TRCTRAP) { 406 /* 407 * XXX not quite right, since this may be for a 408 * multiple fault in user mode. 409 */ 410 MAKEMPSAFE(have_mplock); 411 kprintf("kernel trap %d with interrupts disabled\n", 412 type); 413 } 414 cpu_enable_intr(); 415 } 416 417 type = frame->tf_trapno; 418 code = frame->tf_err; 419 420 if (ISPL(frame->tf_cs) == SEL_UPL) { 421 /* user trap */ 422 423 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 424 frame->tf_trapno, eva); 425 426 userenter(td, p); 427 428 sticks = (int)td->td_sticks; 429 KASSERT(lp->lwp_md.md_regs == frame, 430 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 431 432 switch (type) { 433 case T_PRIVINFLT: /* privileged instruction fault */ 434 ucode = ILL_PRVOPC; 435 i = SIGILL; 436 break; 437 438 case T_BPTFLT: /* bpt instruction fault */ 439 case T_TRCTRAP: /* trace trap */ 440 frame->tf_rflags &= ~PSL_T; 441 ucode = TRAP_TRACE; 442 i = SIGTRAP; 443 break; 444 445 case T_ARITHTRAP: /* arithmetic trap */ 446 ucode = code; 447 i = SIGFPE; 448 #if 0 449 #if JG 450 ucode = fputrap(); 451 #else 452 ucode = code; 453 #endif 454 i = SIGFPE; 455 #endif 456 break; 457 458 case T_ASTFLT: /* Allow process switch */ 459 mycpu->gd_cnt.v_soft++; 460 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 461 atomic_clear_int(&mycpu->gd_reqflags, 462 RQF_AST_OWEUPC); 463 addupc_task(p, p->p_prof.pr_addr, 464 p->p_prof.pr_ticks); 465 } 466 goto out; 467 468 case T_PROTFLT: /* general protection fault */ 469 i = SIGBUS; 470 ucode = BUS_OBJERR; 471 break; 472 case T_SEGNPFLT: /* segment not present fault */ 473 i = SIGBUS; 474 ucode = BUS_ADRERR; 475 break; 476 case T_TSSFLT: /* invalid TSS fault */ 477 case T_DOUBLEFLT: /* double fault */ 478 i = SIGBUS; 479 ucode = BUS_OBJERR; 480 default: 481 #if 0 482 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/ 483 #endif 484 ucode = BUS_OBJERR; 485 i = SIGBUS; 486 break; 487 488 case T_PAGEFLT: /* page fault */ 489 i = trap_pfault(frame, TRUE); 490 if (frame->tf_rip == 0) 491 kprintf("T_PAGEFLT: Warning %%rip == 0!\n"); 492 if (i == -1) 493 goto out; 494 if (i == 0) 495 goto out; 496 497 #if 0 498 ucode = T_PAGEFLT; 499 #endif 500 if (i == SIGSEGV) 501 ucode = SEGV_MAPERR; 502 else 503 ucode = BUS_ADRERR; 504 break; 505 506 case T_DIVIDE: /* integer divide fault */ 507 ucode = FPE_INTDIV; 508 i = SIGFPE; 509 break; 510 511 case T_NMI: 512 MAKEMPSAFE(have_mplock); 513 /* machine/parity/power fail/"kitchen sink" faults */ 514 if (isa_nmi(code) == 0) { 515 #ifdef DDB 516 /* 517 * NMI can be hooked up to a pushbutton 518 * for debugging. 519 */ 520 if (ddb_on_nmi) { 521 kprintf ("NMI ... going to debugger\n"); 522 kdb_trap(type, 0, frame); 523 } 524 #endif /* DDB */ 525 goto out2; 526 } else if (panic_on_nmi) 527 panic("NMI indicates hardware failure"); 528 break; 529 530 case T_OFLOW: /* integer overflow fault */ 531 ucode = FPE_INTOVF; 532 i = SIGFPE; 533 break; 534 535 case T_BOUND: /* bounds check fault */ 536 ucode = FPE_FLTSUB; 537 i = SIGFPE; 538 break; 539 540 case T_DNA: 541 /* 542 * Virtual kernel intercept - pass the DNA exception 543 * to the virtual kernel if it asked to handle it. 544 * This occurs when the virtual kernel is holding 545 * onto the FP context for a different emulated 546 * process then the one currently running. 547 * 548 * We must still call npxdna() since we may have 549 * saved FP state that the virtual kernel needs 550 * to hand over to a different emulated process. 551 */ 552 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 553 (td->td_pcb->pcb_flags & FP_VIRTFP) 554 ) { 555 npxdna(); 556 break; 557 } 558 559 /* 560 * The kernel may have switched out the FP unit's 561 * state, causing the user process to take a fault 562 * when it tries to use the FP unit. Restore the 563 * state here 564 */ 565 if (npxdna()) 566 goto out; 567 i = SIGFPE; 568 ucode = FPE_FPU_NP_TRAP; 569 break; 570 571 case T_FPOPFLT: /* FPU operand fetch fault */ 572 ucode = ILL_COPROC; 573 i = SIGILL; 574 break; 575 576 case T_XMMFLT: /* SIMD floating-point exception */ 577 ucode = 0; /* XXX */ 578 i = SIGFPE; 579 break; 580 } 581 } else { 582 /* kernel trap */ 583 584 switch (type) { 585 case T_PAGEFLT: /* page fault */ 586 trap_pfault(frame, FALSE); 587 goto out2; 588 589 case T_DNA: 590 /* 591 * The kernel is apparently using fpu for copying. 592 * XXX this should be fatal unless the kernel has 593 * registered such use. 594 */ 595 if (npxdna()) 596 goto out2; 597 break; 598 599 case T_STKFLT: /* stack fault */ 600 break; 601 602 case T_PROTFLT: /* general protection fault */ 603 case T_SEGNPFLT: /* segment not present fault */ 604 /* 605 * Invalid segment selectors and out of bounds 606 * %rip's and %rsp's can be set up in user mode. 607 * This causes a fault in kernel mode when the 608 * kernel tries to return to user mode. We want 609 * to get this fault so that we can fix the 610 * problem here and not have to check all the 611 * selectors and pointers when the user changes 612 * them. 613 */ 614 if (mycpu->gd_intr_nesting_level == 0) { 615 if (td->td_pcb->pcb_onfault) { 616 frame->tf_rip = (register_t) 617 td->td_pcb->pcb_onfault; 618 goto out2; 619 } 620 if (frame->tf_rip == (long)doreti_iret) { 621 frame->tf_rip = (long)doreti_iret_fault; 622 goto out2; 623 } 624 } 625 break; 626 627 case T_TSSFLT: 628 /* 629 * PSL_NT can be set in user mode and isn't cleared 630 * automatically when the kernel is entered. This 631 * causes a TSS fault when the kernel attempts to 632 * `iret' because the TSS link is uninitialized. We 633 * want to get this fault so that we can fix the 634 * problem here and not every time the kernel is 635 * entered. 636 */ 637 if (frame->tf_rflags & PSL_NT) { 638 frame->tf_rflags &= ~PSL_NT; 639 goto out2; 640 } 641 break; 642 643 case T_TRCTRAP: /* trace trap */ 644 #if 0 645 if (frame->tf_rip == (int)IDTVEC(syscall)) { 646 /* 647 * We've just entered system mode via the 648 * syscall lcall. Continue single stepping 649 * silently until the syscall handler has 650 * saved the flags. 651 */ 652 goto out2; 653 } 654 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 655 /* 656 * The syscall handler has now saved the 657 * flags. Stop single stepping it. 658 */ 659 frame->tf_rflags &= ~PSL_T; 660 goto out2; 661 } 662 #endif 663 664 /* 665 * Ignore debug register trace traps due to 666 * accesses in the user's address space, which 667 * can happen under several conditions such as 668 * if a user sets a watchpoint on a buffer and 669 * then passes that buffer to a system call. 670 * We still want to get TRCTRAPS for addresses 671 * in kernel space because that is useful when 672 * debugging the kernel. 673 */ 674 #if JG 675 if (user_dbreg_trap()) { 676 /* 677 * Reset breakpoint bits because the 678 * processor doesn't 679 */ 680 /* XXX check upper bits here */ 681 load_dr6(rdr6() & 0xfffffff0); 682 goto out2; 683 } 684 #endif 685 /* 686 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 687 */ 688 case T_BPTFLT: 689 /* 690 * If DDB is enabled, let it handle the debugger trap. 691 * Otherwise, debugger traps "can't happen". 692 */ 693 ucode = TRAP_BRKPT; 694 #ifdef DDB 695 MAKEMPSAFE(have_mplock); 696 if (kdb_trap(type, 0, frame)) 697 goto out2; 698 #endif 699 break; 700 701 case T_NMI: 702 MAKEMPSAFE(have_mplock); 703 /* machine/parity/power fail/"kitchen sink" faults */ 704 #if NISA > 0 705 if (isa_nmi(code) == 0) { 706 #ifdef DDB 707 /* 708 * NMI can be hooked up to a pushbutton 709 * for debugging. 710 */ 711 if (ddb_on_nmi) { 712 kprintf ("NMI ... going to debugger\n"); 713 kdb_trap(type, 0, frame); 714 } 715 #endif /* DDB */ 716 goto out2; 717 } else if (panic_on_nmi == 0) 718 goto out2; 719 /* FALL THROUGH */ 720 #endif /* NISA > 0 */ 721 } 722 MAKEMPSAFE(have_mplock); 723 trap_fatal(frame, 0); 724 goto out2; 725 } 726 727 /* 728 * Virtual kernel intercept - if the fault is directly related to a 729 * VM context managed by a virtual kernel then let the virtual kernel 730 * handle it. 731 */ 732 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 733 vkernel_trap(lp, frame); 734 goto out; 735 } 736 737 /* 738 * Translate fault for emulators (e.g. Linux) 739 */ 740 if (*p->p_sysent->sv_transtrap) 741 i = (*p->p_sysent->sv_transtrap)(i, type); 742 743 MAKEMPSAFE(have_mplock); 744 trapsignal(lp, i, ucode); 745 746 #ifdef DEBUG 747 if (type <= MAX_TRAP_MSG) { 748 uprintf("fatal process exception: %s", 749 trap_msg[type]); 750 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 751 uprintf(", fault VA = 0x%lx", frame->tf_addr); 752 uprintf("\n"); 753 } 754 #endif 755 756 out: 757 userret(lp, frame, sticks); 758 userexit(lp); 759 out2: ; 760 #ifdef SMP 761 if (have_mplock) 762 rel_mplock(); 763 #endif 764 if (p != NULL && lp != NULL) 765 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 766 #ifdef INVARIANTS 767 KASSERT(crit_count == td->td_critcount, 768 ("trap: critical section count mismatch! %d/%d", 769 crit_count, td->td_pri)); 770 KASSERT(curstop == td->td_toks_stop, 771 ("trap: extra tokens held after trap! %ld/%ld", 772 curstop - &td->td_toks_base, 773 td->td_toks_stop - &td->td_toks_base)); 774 #endif 775 } 776 777 static int 778 trap_pfault(struct trapframe *frame, int usermode) 779 { 780 vm_offset_t va; 781 struct vmspace *vm = NULL; 782 vm_map_t map; 783 int rv = 0; 784 int fault_flags; 785 vm_prot_t ftype; 786 thread_t td = curthread; 787 struct lwp *lp = td->td_lwp; 788 struct proc *p; 789 790 va = trunc_page(frame->tf_addr); 791 if (va >= VM_MIN_KERNEL_ADDRESS) { 792 /* 793 * Don't allow user-mode faults in kernel address space. 794 */ 795 if (usermode) { 796 fault_flags = -1; 797 ftype = -1; 798 goto nogo; 799 } 800 801 map = &kernel_map; 802 } else { 803 /* 804 * This is a fault on non-kernel virtual memory. 805 * vm is initialized above to NULL. If curproc is NULL 806 * or curproc->p_vmspace is NULL the fault is fatal. 807 */ 808 if (lp != NULL) 809 vm = lp->lwp_vmspace; 810 811 if (vm == NULL) { 812 fault_flags = -1; 813 ftype = -1; 814 goto nogo; 815 } 816 817 map = &vm->vm_map; 818 } 819 820 /* 821 * PGEX_I is defined only if the execute disable bit capability is 822 * supported and enabled. 823 */ 824 if (frame->tf_err & PGEX_W) 825 ftype = VM_PROT_WRITE; 826 #if JG 827 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 828 ftype = VM_PROT_EXECUTE; 829 #endif 830 else 831 ftype = VM_PROT_READ; 832 833 if (map != &kernel_map) { 834 /* 835 * Keep swapout from messing with us during this 836 * critical time. 837 */ 838 PHOLD(lp->lwp_proc); 839 840 /* 841 * Issue fault 842 */ 843 fault_flags = 0; 844 if (usermode) 845 fault_flags |= VM_FAULT_BURST; 846 if (ftype & VM_PROT_WRITE) 847 fault_flags |= VM_FAULT_DIRTY; 848 else 849 fault_flags |= VM_FAULT_NORMAL; 850 rv = vm_fault(map, va, ftype, fault_flags); 851 852 PRELE(lp->lwp_proc); 853 } else { 854 /* 855 * Don't have to worry about process locking or stacks 856 * in the kernel. 857 */ 858 fault_flags = VM_FAULT_NORMAL; 859 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 860 } 861 862 if (rv == KERN_SUCCESS) 863 return (0); 864 nogo: 865 if (!usermode) { 866 if (td->td_gd->gd_intr_nesting_level == 0 && 867 td->td_pcb->pcb_onfault) { 868 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 869 return (0); 870 } 871 trap_fatal(frame, frame->tf_addr); 872 return (-1); 873 } 874 875 /* 876 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 877 * kludge is needed to pass the fault address to signal handlers. 878 */ 879 p = td->td_proc; 880 if (td->td_lwp->lwp_vkernel == NULL) { 881 if (bootverbose) 882 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p " 883 "pid=%d p_comm=%s\n", 884 ftype, fault_flags, 885 (void *)frame->tf_addr, 886 (void *)frame->tf_rip, 887 p->p_pid, p->p_comm); 888 if (ddb_on_seg_fault) 889 Debugger("ddb_on_seg_fault"); 890 } 891 /* Debugger("seg-fault"); */ 892 893 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 894 } 895 896 static void 897 trap_fatal(struct trapframe *frame, vm_offset_t eva) 898 { 899 int code, ss; 900 u_int type; 901 long rsp; 902 struct soft_segment_descriptor softseg; 903 char *msg; 904 905 code = frame->tf_err; 906 type = frame->tf_trapno; 907 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 908 909 if (type <= MAX_TRAP_MSG) 910 msg = trap_msg[type]; 911 else 912 msg = "UNKNOWN"; 913 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 914 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 915 #ifdef SMP 916 /* three separate prints in case of a trap on an unmapped page */ 917 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 918 kprintf("lapic->id = %08x\n", lapic->id); 919 #endif 920 if (type == T_PAGEFLT) { 921 kprintf("fault virtual address = 0x%lx\n", eva); 922 kprintf("fault code = %s %s %s, %s\n", 923 code & PGEX_U ? "user" : "supervisor", 924 code & PGEX_W ? "write" : "read", 925 code & PGEX_I ? "instruction" : "data", 926 code & PGEX_P ? "protection violation" : "page not present"); 927 } 928 kprintf("instruction pointer = 0x%lx:0x%lx\n", 929 frame->tf_cs & 0xffff, frame->tf_rip); 930 if (ISPL(frame->tf_cs) == SEL_UPL) { 931 ss = frame->tf_ss & 0xffff; 932 rsp = frame->tf_rsp; 933 } else { 934 ss = GSEL(GDATA_SEL, SEL_KPL); 935 rsp = (long)&frame->tf_rsp; 936 } 937 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 938 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 939 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 940 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 941 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 942 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 943 softseg.ssd_gran); 944 kprintf("processor eflags = "); 945 if (frame->tf_rflags & PSL_T) 946 kprintf("trace trap, "); 947 if (frame->tf_rflags & PSL_I) 948 kprintf("interrupt enabled, "); 949 if (frame->tf_rflags & PSL_NT) 950 kprintf("nested task, "); 951 if (frame->tf_rflags & PSL_RF) 952 kprintf("resume, "); 953 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 954 kprintf("current process = "); 955 if (curproc) { 956 kprintf("%lu\n", 957 (u_long)curproc->p_pid); 958 } else { 959 kprintf("Idle\n"); 960 } 961 kprintf("current thread = pri %d ", curthread->td_pri); 962 if (curthread->td_critcount) 963 kprintf("(CRIT)"); 964 kprintf("\n"); 965 966 #ifdef DDB 967 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 968 return; 969 #endif 970 kprintf("trap number = %d\n", type); 971 if (type <= MAX_TRAP_MSG) 972 panic("%s", trap_msg[type]); 973 else 974 panic("unknown/reserved trap"); 975 } 976 977 /* 978 * Double fault handler. Called when a fault occurs while writing 979 * a frame for a trap/exception onto the stack. This usually occurs 980 * when the stack overflows (such is the case with infinite recursion, 981 * for example). 982 */ 983 static __inline 984 int 985 in_kstack_guard(register_t rptr) 986 { 987 thread_t td = curthread; 988 989 if ((char *)rptr >= td->td_kstack && 990 (char *)rptr < td->td_kstack + PAGE_SIZE) { 991 return 1; 992 } 993 return 0; 994 } 995 996 void 997 dblfault_handler(struct trapframe *frame) 998 { 999 thread_t td = curthread; 1000 1001 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1002 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1003 if (in_kstack_guard(frame->tf_rsp)) 1004 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1005 if (in_kstack_guard(frame->tf_rbp)) 1006 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1007 } else { 1008 kprintf("DOUBLE FAULT\n"); 1009 } 1010 kprintf("\nFatal double fault\n"); 1011 kprintf("rip = 0x%lx\n", frame->tf_rip); 1012 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1013 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1014 #ifdef SMP 1015 /* three separate prints in case of a trap on an unmapped page */ 1016 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1017 kprintf("lapic->id = %08x\n", lapic->id); 1018 #endif 1019 panic("double fault"); 1020 } 1021 1022 /* 1023 * syscall2 - MP aware system call request C handler 1024 * 1025 * A system call is essentially treated as a trap except that the 1026 * MP lock is not held on entry or return. We are responsible for 1027 * obtaining the MP lock if necessary and for handling ASTs 1028 * (e.g. a task switch) prior to return. 1029 * 1030 * MPSAFE 1031 */ 1032 void 1033 syscall2(struct trapframe *frame) 1034 { 1035 struct thread *td = curthread; 1036 struct proc *p = td->td_proc; 1037 struct lwp *lp = td->td_lwp; 1038 caddr_t params; 1039 struct sysent *callp; 1040 register_t orig_tf_rflags; 1041 int sticks; 1042 int error; 1043 int narg; 1044 #ifdef INVARIANTS 1045 int crit_count = td->td_critcount; 1046 #endif 1047 #ifdef SMP 1048 int have_mplock = 0; 1049 #endif 1050 register_t *argp; 1051 u_int code; 1052 int reg, regcnt; 1053 union sysunion args; 1054 register_t *argsdst; 1055 1056 mycpu->gd_cnt.v_syscall++; 1057 1058 #ifdef DIAGNOSTIC 1059 if (ISPL(frame->tf_cs) != SEL_UPL) { 1060 get_mplock(); 1061 panic("syscall"); 1062 /* NOT REACHED */ 1063 } 1064 #endif 1065 1066 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1067 frame->tf_rax); 1068 1069 userenter(td, p); /* lazy raise our priority */ 1070 1071 reg = 0; 1072 regcnt = 6; 1073 /* 1074 * Misc 1075 */ 1076 sticks = (int)td->td_sticks; 1077 orig_tf_rflags = frame->tf_rflags; 1078 1079 /* 1080 * Virtual kernel intercept - if a VM context managed by a virtual 1081 * kernel issues a system call the virtual kernel handles it, not us. 1082 * Restore the virtual kernel context and return from its system 1083 * call. The current frame is copied out to the virtual kernel. 1084 */ 1085 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1086 vkernel_trap(lp, frame); 1087 error = EJUSTRETURN; 1088 goto out; 1089 } 1090 1091 /* 1092 * Get the system call parameters and account for time 1093 */ 1094 KASSERT(lp->lwp_md.md_regs == frame, 1095 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1096 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1097 code = frame->tf_rax; 1098 1099 if (p->p_sysent->sv_prepsyscall) { 1100 (*p->p_sysent->sv_prepsyscall)( 1101 frame, (int *)(&args.nosys.sysmsg + 1), 1102 &code, ¶ms); 1103 } else { 1104 if (code == SYS_syscall || code == SYS___syscall) { 1105 code = frame->tf_rdi; 1106 reg++; 1107 regcnt--; 1108 } 1109 } 1110 1111 if (p->p_sysent->sv_mask) 1112 code &= p->p_sysent->sv_mask; 1113 1114 if (code >= p->p_sysent->sv_size) 1115 callp = &p->p_sysent->sv_table[0]; 1116 else 1117 callp = &p->p_sysent->sv_table[code]; 1118 1119 narg = callp->sy_narg & SYF_ARGMASK; 1120 1121 /* 1122 * On x86_64 we get up to six arguments in registers. The rest are 1123 * on the stack. The first six members of 'struct trapframe' happen 1124 * to be the registers used to pass arguments, in exactly the right 1125 * order. 1126 */ 1127 argp = &frame->tf_rdi; 1128 argp += reg; 1129 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1130 /* 1131 * JG can we overflow the space pointed to by 'argsdst' 1132 * either with 'bcopy' or with 'copyin'? 1133 */ 1134 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1135 /* 1136 * copyin is MP aware, but the tracing code is not 1137 */ 1138 if (narg > regcnt) { 1139 KASSERT(params != NULL, ("copyin args with no params!")); 1140 error = copyin(params, &argsdst[regcnt], 1141 (narg - regcnt) * sizeof(register_t)); 1142 if (error) { 1143 #ifdef KTRACE 1144 if (KTRPOINT(td, KTR_SYSCALL)) { 1145 MAKEMPSAFE(have_mplock); 1146 1147 ktrsyscall(lp, code, narg, 1148 (void *)(&args.nosys.sysmsg + 1)); 1149 } 1150 #endif 1151 goto bad; 1152 } 1153 } 1154 1155 #ifdef KTRACE 1156 if (KTRPOINT(td, KTR_SYSCALL)) { 1157 MAKEMPSAFE(have_mplock); 1158 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1159 } 1160 #endif 1161 1162 /* 1163 * Default return value is 0 (will be copied to %rax). Double-value 1164 * returns use %rax and %rdx. %rdx is left unchanged for system 1165 * calls which return only one result. 1166 */ 1167 args.sysmsg_fds[0] = 0; 1168 args.sysmsg_fds[1] = frame->tf_rdx; 1169 1170 /* 1171 * The syscall might manipulate the trap frame. If it does it 1172 * will probably return EJUSTRETURN. 1173 */ 1174 args.sysmsg_frame = frame; 1175 1176 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1177 1178 /* 1179 * NOTE: All system calls run MPSAFE now. The system call itself 1180 * is responsible for getting the MP lock. 1181 */ 1182 error = (*callp->sy_call)(&args); 1183 1184 out: 1185 /* 1186 * MP SAFE (we may or may not have the MP lock at this point) 1187 */ 1188 //kprintf("SYSMSG %d ", error); 1189 switch (error) { 1190 case 0: 1191 /* 1192 * Reinitialize proc pointer `p' as it may be different 1193 * if this is a child returning from fork syscall. 1194 */ 1195 p = curproc; 1196 lp = curthread->td_lwp; 1197 frame->tf_rax = args.sysmsg_fds[0]; 1198 frame->tf_rdx = args.sysmsg_fds[1]; 1199 frame->tf_rflags &= ~PSL_C; 1200 break; 1201 case ERESTART: 1202 /* 1203 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1204 * We have to do a full context restore so that %r10 1205 * (which was holding the value of %rcx) is restored for 1206 * the next iteration. 1207 */ 1208 frame->tf_rip -= frame->tf_err; 1209 frame->tf_r10 = frame->tf_rcx; 1210 break; 1211 case EJUSTRETURN: 1212 break; 1213 case EASYNC: 1214 panic("Unexpected EASYNC return value (for now)"); 1215 default: 1216 bad: 1217 if (p->p_sysent->sv_errsize) { 1218 if (error >= p->p_sysent->sv_errsize) 1219 error = -1; /* XXX */ 1220 else 1221 error = p->p_sysent->sv_errtbl[error]; 1222 } 1223 frame->tf_rax = error; 1224 frame->tf_rflags |= PSL_C; 1225 break; 1226 } 1227 1228 /* 1229 * Traced syscall. trapsignal() is not MP aware. 1230 */ 1231 if (orig_tf_rflags & PSL_T) { 1232 MAKEMPSAFE(have_mplock); 1233 frame->tf_rflags &= ~PSL_T; 1234 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1235 } 1236 1237 /* 1238 * Handle reschedule and other end-of-syscall issues 1239 */ 1240 userret(lp, frame, sticks); 1241 1242 #ifdef KTRACE 1243 if (KTRPOINT(td, KTR_SYSRET)) { 1244 MAKEMPSAFE(have_mplock); 1245 ktrsysret(lp, code, error, args.sysmsg_result); 1246 } 1247 #endif 1248 1249 /* 1250 * This works because errno is findable through the 1251 * register set. If we ever support an emulation where this 1252 * is not the case, this code will need to be revisited. 1253 */ 1254 STOPEVENT(p, S_SCX, code); 1255 1256 userexit(lp); 1257 #ifdef SMP 1258 /* 1259 * Release the MP lock if we had to get it 1260 */ 1261 if (have_mplock) 1262 rel_mplock(); 1263 #endif 1264 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1265 #ifdef INVARIANTS 1266 KASSERT(crit_count == td->td_critcount, 1267 ("syscall: critical section count mismatch! %d/%d", 1268 crit_count, td->td_pri)); 1269 KASSERT(&td->td_toks_base == td->td_toks_stop, 1270 ("syscall: extra tokens held after trap! %ld", 1271 td->td_toks_stop - &td->td_toks_base)); 1272 #endif 1273 } 1274 1275 /* 1276 * NOTE: mplock not held at any point 1277 */ 1278 void 1279 fork_return(struct lwp *lp, struct trapframe *frame) 1280 { 1281 frame->tf_rax = 0; /* Child returns zero */ 1282 frame->tf_rflags &= ~PSL_C; /* success */ 1283 frame->tf_rdx = 1; 1284 1285 generic_lwp_return(lp, frame); 1286 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1287 } 1288 1289 /* 1290 * Simplified back end of syscall(), used when returning from fork() 1291 * directly into user mode. 1292 * 1293 * This code will return back into the fork trampoline code which then 1294 * runs doreti. 1295 * 1296 * NOTE: The mplock is not held at any point. 1297 */ 1298 void 1299 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1300 { 1301 struct proc *p = lp->lwp_proc; 1302 1303 /* 1304 * Newly forked processes are given a kernel priority. We have to 1305 * adjust the priority to a normal user priority and fake entry 1306 * into the kernel (call userenter()) to install a passive release 1307 * function just in case userret() decides to stop the process. This 1308 * can occur when ^Z races a fork. If we do not install the passive 1309 * release function the current process designation will not be 1310 * released when the thread goes to sleep. 1311 */ 1312 lwkt_setpri_self(TDPRI_USER_NORM); 1313 userenter(lp->lwp_thread, p); 1314 userret(lp, frame, 0); 1315 #ifdef KTRACE 1316 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1317 ktrsysret(lp, SYS_fork, 0, 0); 1318 #endif 1319 p->p_flag |= P_PASSIVE_ACQ; 1320 userexit(lp); 1321 p->p_flag &= ~P_PASSIVE_ACQ; 1322 } 1323 1324 /* 1325 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1326 * fault (which is then passed back to the virtual kernel) if an attempt is 1327 * made to use the FP unit. 1328 * 1329 * XXX this is a fairly big hack. 1330 */ 1331 void 1332 set_vkernel_fp(struct trapframe *frame) 1333 { 1334 struct thread *td = curthread; 1335 1336 if (frame->tf_xflags & PGEX_FPFAULT) { 1337 td->td_pcb->pcb_flags |= FP_VIRTFP; 1338 if (mdcpu->gd_npxthread == td) 1339 npxexit(); 1340 } else { 1341 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1342 } 1343 } 1344 1345 /* 1346 * Called from vkernel_trap() to fixup the vkernel's syscall 1347 * frame for vmspace_ctl() return. 1348 */ 1349 void 1350 cpu_vkernel_trap(struct trapframe *frame, int error) 1351 { 1352 frame->tf_rax = error; 1353 if (error) 1354 frame->tf_rflags |= PSL_C; 1355 else 1356 frame->tf_rflags &= ~PSL_C; 1357 } 1358