subr_syscall.c revision 65781
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 * $FreeBSD: head/sys/kern/subr_trap.c 65781 2000-09-12 18:41:56Z bde $ 39 */ 40 41/* 42 * 386 Trap and System call handling 43 */ 44 45#include "opt_cpu.h" 46#include "opt_ddb.h" 47#include "opt_ktrace.h" 48#include "opt_clock.h" 49#include "opt_trap.h" 50 51#include <sys/param.h> 52#include <sys/bus.h> 53#include <sys/systm.h> 54#include <sys/proc.h> 55#include <sys/pioctl.h> 56#include <sys/kernel.h> 57#include <sys/ktr.h> 58#include <sys/resourcevar.h> 59#include <sys/signalvar.h> 60#include <sys/syscall.h> 61#include <sys/sysctl.h> 62#include <sys/sysent.h> 63#include <sys/uio.h> 64#include <sys/vmmeter.h> 65#ifdef KTRACE 66#include <sys/ktrace.h> 67#endif 68 69#include <vm/vm.h> 70#include <vm/vm_param.h> 71#include <sys/lock.h> 72#include <vm/pmap.h> 73#include <vm/vm_kern.h> 74#include <vm/vm_map.h> 75#include <vm/vm_page.h> 76#include <vm/vm_extern.h> 77 78#include <machine/cpu.h> 79#include <machine/ipl.h> 80#include <machine/md_var.h> 81#include <machine/mutex.h> 82#include <machine/pcb.h> 83#ifdef SMP 84#include <machine/smp.h> 85#endif 86#include <machine/tss.h> 87 88#include <i386/isa/icu.h> 89#include <i386/isa/intr_machdep.h> 90 91#ifdef POWERFAIL_NMI 92#include <sys/syslog.h> 93#include <machine/clock.h> 94#endif 95 96#include <machine/vm86.h> 97 98#include <ddb/ddb.h> 99 100#include "isa.h" 101#include "npx.h" 102 103#include <sys/sysctl.h> 104 105int (*pmath_emulate) __P((struct trapframe *)); 106 107extern void trap __P((struct trapframe frame)); 108extern int trapwrite __P((unsigned addr)); 109extern void syscall2 __P((struct trapframe frame)); 110extern void ast __P((struct trapframe frame)); 111 112static int trap_pfault __P((struct trapframe *, int, vm_offset_t)); 113static void trap_fatal __P((struct trapframe *, vm_offset_t)); 114void dblfault_handler __P((void)); 115 116extern inthand_t IDTVEC(syscall); 117 118#define MAX_TRAP_MSG 28 119static char *trap_msg[] = { 120 "", /* 0 unused */ 121 "privileged instruction fault", /* 1 T_PRIVINFLT */ 122 "", /* 2 unused */ 123 "breakpoint instruction fault", /* 3 T_BPTFLT */ 124 "", /* 4 unused */ 125 "", /* 5 unused */ 126 "arithmetic trap", /* 6 T_ARITHTRAP */ 127 "system forced exception", /* 7 T_ASTFLT */ 128 "", /* 8 unused */ 129 "general protection fault", /* 9 T_PROTFLT */ 130 "trace trap", /* 10 T_TRCTRAP */ 131 "", /* 11 unused */ 132 "page fault", /* 12 T_PAGEFLT */ 133 "", /* 13 unused */ 134 "alignment fault", /* 14 T_ALIGNFLT */ 135 "", /* 15 unused */ 136 "", /* 16 unused */ 137 "", /* 17 unused */ 138 "integer divide fault", /* 18 T_DIVIDE */ 139 "non-maskable interrupt trap", /* 19 T_NMI */ 140 "overflow trap", /* 20 T_OFLOW */ 141 "FPU bounds check fault", /* 21 T_BOUND */ 142 "FPU device not available", /* 22 T_DNA */ 143 "double fault", /* 23 T_DOUBLEFLT */ 144 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 145 "invalid TSS fault", /* 25 T_TSSFLT */ 146 "segment not present fault", /* 26 T_SEGNPFLT */ 147 "stack fault", /* 27 T_STKFLT */ 148 "machine check trap", /* 28 T_MCHK */ 149}; 150 151static __inline int userret __P((struct proc *p, struct trapframe *frame, 152 u_quad_t oticks, int have_giant)); 153 154#if defined(I586_CPU) && !defined(NO_F00F_HACK) 155extern int has_f00f_bug; 156#endif 157 158#ifdef DDB 159static int ddb_on_nmi = 1; 160SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 161 &ddb_on_nmi, 0, "Go to DDB on NMI"); 162#endif 163static int panic_on_nmi = 1; 164SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 165 &panic_on_nmi, 0, "Panic on NMI"); 166 167static __inline int 168userret(p, frame, oticks, have_giant) 169 struct proc *p; 170 struct trapframe *frame; 171 u_quad_t oticks; 172 int have_giant; 173{ 174 int sig, s; 175 176 while ((sig = CURSIG(p)) != 0) { 177 if (have_giant == 0) { 178 mtx_enter(&Giant, MTX_DEF); 179 have_giant = 1; 180 } 181 postsig(sig); 182 } 183 184 p->p_priority = p->p_usrpri; 185 if (resched_wanted()) { 186 /* 187 * Since we are curproc, clock will normally just change 188 * our priority without moving us from one queue to another 189 * (since the running process is not on a queue.) 190 * If that happened after we setrunqueue ourselves but before we 191 * mi_switch()'ed, we might not be on the queue indicated by 192 * our priority. 193 */ 194 s = splhigh(); 195 mtx_enter(&sched_lock, MTX_SPIN); 196 setrunqueue(p); 197 p->p_stats->p_ru.ru_nivcsw++; 198 mi_switch(); 199 mtx_exit(&sched_lock, MTX_SPIN); 200 splx(s); 201 while ((sig = CURSIG(p)) != 0) { 202 if (have_giant == 0) { 203 mtx_enter(&Giant, MTX_DEF); 204 have_giant = 1; 205 } 206 postsig(sig); 207 } 208 } 209 /* 210 * Charge system time if profiling. 211 */ 212 if (p->p_flag & P_PROFIL) { 213 if (have_giant == 0) { 214 mtx_enter(&Giant, MTX_DEF); 215 have_giant = 1; 216 } 217 addupc_task(p, frame->tf_eip, 218 (u_int)(p->p_sticks - oticks) * psratio); 219 } 220 curpriority = p->p_priority; 221 return(have_giant); 222} 223 224/* 225 * Exception, fault, and trap interface to the FreeBSD kernel. 226 * This common code is called from assembly language IDT gate entry 227 * routines that prepare a suitable stack frame, and restore this 228 * frame after the exception has been processed. 229 */ 230 231void 232trap(frame) 233 struct trapframe frame; 234{ 235 struct proc *p = curproc; 236 u_quad_t sticks = 0; 237 int i = 0, ucode = 0, type, code; 238 vm_offset_t eva; 239#ifdef POWERFAIL_NMI 240 static int lastalert = 0; 241#endif 242 243 atomic_add_int(&cnt.v_trap, 1); 244 245 if ((frame.tf_eflags & PSL_I) == 0) { 246 /* 247 * Buggy application or kernel code has disabled 248 * interrupts and then trapped. Enabling interrupts 249 * now is wrong, but it is better than running with 250 * interrupts disabled until they are accidentally 251 * enabled later. XXX Consider whether is this still 252 * correct. 253 */ 254 type = frame.tf_trapno; 255 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) 256 printf( 257 "pid %ld (%s): trap %d with interrupts disabled\n", 258 (long)curproc->p_pid, curproc->p_comm, type); 259 else if (type != T_BPTFLT && type != T_TRCTRAP) 260 /* 261 * XXX not quite right, since this may be for a 262 * multiple fault in user mode. 263 */ 264 printf("kernel trap %d with interrupts disabled\n", 265 type); 266 enable_intr(); 267 } 268 269 eva = 0; 270 if (frame.tf_trapno == T_PAGEFLT) { 271 /* 272 * For some Cyrix CPUs, %cr2 is clobbered by 273 * interrupts. This problem is worked around by using 274 * an interrupt gate for the pagefault handler. We 275 * are finally ready to read %cr2 and then must 276 * reenable interrupts. 277 */ 278 eva = rcr2(); 279 enable_intr(); 280 } 281 282 if (p != NULL || !cold) 283 mtx_enter(&Giant, MTX_DEF); 284 285#if defined(I586_CPU) && !defined(NO_F00F_HACK) 286restart: 287#endif 288 289 type = frame.tf_trapno; 290 code = frame.tf_err; 291 292 if ((ISPL(frame.tf_cs) == SEL_UPL) || 293 ((frame.tf_eflags & PSL_VM) && !in_vm86call)) { 294 /* user trap */ 295 296 sticks = p->p_sticks; 297 p->p_md.md_regs = &frame; 298 299 switch (type) { 300 case T_PRIVINFLT: /* privileged instruction fault */ 301 ucode = type; 302 i = SIGILL; 303 break; 304 305 case T_BPTFLT: /* bpt instruction fault */ 306 case T_TRCTRAP: /* trace trap */ 307 frame.tf_eflags &= ~PSL_T; 308 i = SIGTRAP; 309 break; 310 311 case T_ARITHTRAP: /* arithmetic trap */ 312 ucode = code; 313 i = SIGFPE; 314 break; 315 316 /* 317 * The following two traps can happen in 318 * vm86 mode, and, if so, we want to handle 319 * them specially. 320 */ 321 case T_PROTFLT: /* general protection fault */ 322 case T_STKFLT: /* stack fault */ 323 if (frame.tf_eflags & PSL_VM) { 324 i = vm86_emulate((struct vm86frame *)&frame); 325 if (i == 0) 326 goto user; 327 break; 328 } 329 /* FALL THROUGH */ 330 331 case T_SEGNPFLT: /* segment not present fault */ 332 case T_TSSFLT: /* invalid TSS fault */ 333 case T_DOUBLEFLT: /* double fault */ 334 default: 335 ucode = code + BUS_SEGM_FAULT ; 336 i = SIGBUS; 337 break; 338 339 case T_PAGEFLT: /* page fault */ 340 i = trap_pfault(&frame, TRUE, eva); 341#if defined(I586_CPU) && !defined(NO_F00F_HACK) 342 if (i == -2) { 343 /* 344 * f00f hack workaround has triggered, treat 345 * as illegal instruction not page fault. 346 */ 347 frame.tf_trapno = T_PRIVINFLT; 348 goto restart; 349 } 350#endif 351 if (i == -1) 352 goto out; 353 if (i == 0) 354 goto user; 355 356 ucode = T_PAGEFLT; 357 break; 358 359 case T_DIVIDE: /* integer divide fault */ 360 ucode = FPE_INTDIV; 361 i = SIGFPE; 362 break; 363 364#if NISA > 0 365 case T_NMI: 366#ifdef POWERFAIL_NMI 367#ifndef TIMER_FREQ 368# define TIMER_FREQ 1193182 369#endif 370 if (time_second - lastalert > 10) { 371 log(LOG_WARNING, "NMI: power fail\n"); 372 sysbeep(TIMER_FREQ/880, hz); 373 lastalert = time_second; 374 } 375 goto out; 376#else /* !POWERFAIL_NMI */ 377 /* machine/parity/power fail/"kitchen sink" faults */ 378 if (isa_nmi(code) == 0) { 379#ifdef DDB 380 /* 381 * NMI can be hooked up to a pushbutton 382 * for debugging. 383 */ 384 if (ddb_on_nmi) { 385 printf ("NMI ... going to debugger\n"); 386 kdb_trap (type, 0, &frame); 387 } 388#endif /* DDB */ 389 goto out; 390 } else if (panic_on_nmi) 391 panic("NMI indicates hardware failure"); 392 break; 393#endif /* POWERFAIL_NMI */ 394#endif /* NISA > 0 */ 395 396 case T_OFLOW: /* integer overflow fault */ 397 ucode = FPE_INTOVF; 398 i = SIGFPE; 399 break; 400 401 case T_BOUND: /* bounds check fault */ 402 ucode = FPE_FLTSUB; 403 i = SIGFPE; 404 break; 405 406 case T_DNA: 407#if NNPX > 0 408 /* transparent fault (due to context switch "late") */ 409 if (npxdna()) 410 goto out; 411#endif 412 if (!pmath_emulate) { 413 i = SIGFPE; 414 ucode = FPE_FPU_NP_TRAP; 415 break; 416 } 417 i = (*pmath_emulate)(&frame); 418 if (i == 0) { 419 if (!(frame.tf_eflags & PSL_T)) 420 goto out; 421 frame.tf_eflags &= ~PSL_T; 422 i = SIGTRAP; 423 } 424 /* else ucode = emulator_only_knows() XXX */ 425 break; 426 427 case T_FPOPFLT: /* FPU operand fetch fault */ 428 ucode = T_FPOPFLT; 429 i = SIGILL; 430 break; 431 } 432 } else { 433 /* kernel trap */ 434 435 switch (type) { 436 case T_PAGEFLT: /* page fault */ 437 (void) trap_pfault(&frame, FALSE, eva); 438 goto out; 439 440 case T_DNA: 441#if NNPX > 0 442 /* 443 * The kernel is apparently using npx for copying. 444 * XXX this should be fatal unless the kernel has 445 * registered such use. 446 */ 447 if (npxdna()) 448 goto out; 449#endif 450 break; 451 452 /* 453 * The following two traps can happen in 454 * vm86 mode, and, if so, we want to handle 455 * them specially. 456 */ 457 case T_PROTFLT: /* general protection fault */ 458 case T_STKFLT: /* stack fault */ 459 if (frame.tf_eflags & PSL_VM) { 460 i = vm86_emulate((struct vm86frame *)&frame); 461 if (i != 0) 462 /* 463 * returns to original process 464 */ 465 vm86_trap((struct vm86frame *)&frame); 466 goto out; 467 } 468 /* FALL THROUGH */ 469 470 case T_SEGNPFLT: /* segment not present fault */ 471 if (in_vm86call) 472 break; 473 474 if (intr_nesting_level != 0) 475 break; 476 477 /* 478 * Invalid %fs's and %gs's can be created using 479 * procfs or PT_SETREGS or by invalidating the 480 * underlying LDT entry. This causes a fault 481 * in kernel mode when the kernel attempts to 482 * switch contexts. Lose the bad context 483 * (XXX) so that we can continue, and generate 484 * a signal. 485 */ 486 if (frame.tf_eip == (int)cpu_switch_load_gs) { 487 curpcb->pcb_gs = 0; 488 psignal(p, SIGBUS); 489 goto out; 490 } 491 492 /* 493 * Invalid segment selectors and out of bounds 494 * %eip's and %esp's can be set up in user mode. 495 * This causes a fault in kernel mode when the 496 * kernel tries to return to user mode. We want 497 * to get this fault so that we can fix the 498 * problem here and not have to check all the 499 * selectors and pointers when the user changes 500 * them. 501 */ 502 if (frame.tf_eip == (int)doreti_iret) { 503 frame.tf_eip = (int)doreti_iret_fault; 504 goto out; 505 } 506 if (frame.tf_eip == (int)doreti_popl_ds) { 507 frame.tf_eip = (int)doreti_popl_ds_fault; 508 goto out; 509 } 510 if (frame.tf_eip == (int)doreti_popl_es) { 511 frame.tf_eip = (int)doreti_popl_es_fault; 512 goto out; 513 } 514 if (frame.tf_eip == (int)doreti_popl_fs) { 515 frame.tf_eip = (int)doreti_popl_fs_fault; 516 goto out; 517 } 518 if (curpcb && curpcb->pcb_onfault) { 519 frame.tf_eip = (int)curpcb->pcb_onfault; 520 goto out; 521 } 522 break; 523 524 case T_TSSFLT: 525 /* 526 * PSL_NT can be set in user mode and isn't cleared 527 * automatically when the kernel is entered. This 528 * causes a TSS fault when the kernel attempts to 529 * `iret' because the TSS link is uninitialized. We 530 * want to get this fault so that we can fix the 531 * problem here and not every time the kernel is 532 * entered. 533 */ 534 if (frame.tf_eflags & PSL_NT) { 535 frame.tf_eflags &= ~PSL_NT; 536 goto out; 537 } 538 break; 539 540 case T_TRCTRAP: /* trace trap */ 541 if (frame.tf_eip == (int)IDTVEC(syscall)) { 542 /* 543 * We've just entered system mode via the 544 * syscall lcall. Continue single stepping 545 * silently until the syscall handler has 546 * saved the flags. 547 */ 548 goto out; 549 } 550 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { 551 /* 552 * The syscall handler has now saved the 553 * flags. Stop single stepping it. 554 */ 555 frame.tf_eflags &= ~PSL_T; 556 goto out; 557 } 558 /* 559 * Ignore debug register trace traps due to 560 * accesses in the user's address space, which 561 * can happen under several conditions such as 562 * if a user sets a watchpoint on a buffer and 563 * then passes that buffer to a system call. 564 * We still want to get TRCTRAPS for addresses 565 * in kernel space because that is useful when 566 * debugging the kernel. 567 */ 568 if (user_dbreg_trap() && !in_vm86call) { 569 /* 570 * Reset breakpoint bits because the 571 * processor doesn't 572 */ 573 load_dr6(rdr6() & 0xfffffff0); 574 goto out; 575 } 576 /* 577 * Fall through (TRCTRAP kernel mode, kernel address) 578 */ 579 case T_BPTFLT: 580 /* 581 * If DDB is enabled, let it handle the debugger trap. 582 * Otherwise, debugger traps "can't happen". 583 */ 584#ifdef DDB 585 if (kdb_trap (type, 0, &frame)) 586 goto out; 587#endif 588 break; 589 590#if NISA > 0 591 case T_NMI: 592#ifdef POWERFAIL_NMI 593 if (time_second - lastalert > 10) { 594 log(LOG_WARNING, "NMI: power fail\n"); 595 sysbeep(TIMER_FREQ/880, hz); 596 lastalert = time_second; 597 } 598 goto out; 599#else /* !POWERFAIL_NMI */ 600 /* machine/parity/power fail/"kitchen sink" faults */ 601 if (isa_nmi(code) == 0) { 602#ifdef DDB 603 /* 604 * NMI can be hooked up to a pushbutton 605 * for debugging. 606 */ 607 if (ddb_on_nmi) { 608 printf ("NMI ... going to debugger\n"); 609 kdb_trap (type, 0, &frame); 610 } 611#endif /* DDB */ 612 goto out; 613 } else if (panic_on_nmi == 0) 614 goto out; 615 /* FALL THROUGH */ 616#endif /* POWERFAIL_NMI */ 617#endif /* NISA > 0 */ 618 } 619 620 trap_fatal(&frame, eva); 621 goto out; 622 } 623 624 /* Translate fault for emulators (e.g. Linux) */ 625 if (*p->p_sysent->sv_transtrap) 626 i = (*p->p_sysent->sv_transtrap)(i, type); 627 628 trapsignal(p, i, ucode); 629 630#ifdef DEBUG 631 if (type <= MAX_TRAP_MSG) { 632 uprintf("fatal process exception: %s", 633 trap_msg[type]); 634 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 635 uprintf(", fault VA = 0x%lx", (u_long)eva); 636 uprintf("\n"); 637 } 638#endif 639 640user: 641 userret(p, &frame, sticks, 1); 642out: 643 if (p != NULL || !cold) 644 mtx_exit(&Giant, MTX_DEF); 645} 646 647#ifdef notyet 648/* 649 * This version doesn't allow a page fault to user space while 650 * in the kernel. The rest of the kernel needs to be made "safe" 651 * before this can be used. I think the only things remaining 652 * to be made safe are the iBCS2 code and the process tracing/ 653 * debugging code. 654 */ 655static int 656trap_pfault(frame, usermode, eva) 657 struct trapframe *frame; 658 int usermode; 659 vm_offset_t eva; 660{ 661 vm_offset_t va; 662 struct vmspace *vm = NULL; 663 vm_map_t map = 0; 664 int rv = 0; 665 vm_prot_t ftype; 666 struct proc *p = curproc; 667 668 if (frame->tf_err & PGEX_W) 669 ftype = VM_PROT_WRITE; 670 else 671 ftype = VM_PROT_READ; 672 673 va = trunc_page(eva); 674 if (va < VM_MIN_KERNEL_ADDRESS) { 675 vm_offset_t v; 676 vm_page_t mpte; 677 678 if (p == NULL || 679 (!usermode && va < VM_MAXUSER_ADDRESS && 680 (intr_nesting_level != 0 || curpcb == NULL || 681 curpcb->pcb_onfault == NULL))) { 682 trap_fatal(frame, eva); 683 return (-1); 684 } 685 686 /* 687 * This is a fault on non-kernel virtual memory. 688 * vm is initialized above to NULL. If curproc is NULL 689 * or curproc->p_vmspace is NULL the fault is fatal. 690 */ 691 vm = p->p_vmspace; 692 if (vm == NULL) 693 goto nogo; 694 695 map = &vm->vm_map; 696 697 /* 698 * Keep swapout from messing with us during this 699 * critical time. 700 */ 701 ++p->p_lock; 702 703 /* 704 * Grow the stack if necessary 705 */ 706 /* grow_stack returns false only if va falls into 707 * a growable stack region and the stack growth 708 * fails. It returns true if va was not within 709 * a growable stack region, or if the stack 710 * growth succeeded. 711 */ 712 if (!grow_stack (p, va)) { 713 rv = KERN_FAILURE; 714 --p->p_lock; 715 goto nogo; 716 } 717 718 /* Fault in the user page: */ 719 rv = vm_fault(map, va, ftype, 720 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 721 : VM_FAULT_NORMAL); 722 723 --p->p_lock; 724 } else { 725 /* 726 * Don't allow user-mode faults in kernel address space. 727 */ 728 if (usermode) 729 goto nogo; 730 731 /* 732 * Since we know that kernel virtual address addresses 733 * always have pte pages mapped, we just have to fault 734 * the page. 735 */ 736 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL); 737 } 738 739 if (rv == KERN_SUCCESS) 740 return (0); 741nogo: 742 if (!usermode) { 743 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 744 frame->tf_eip = (int)curpcb->pcb_onfault; 745 return (0); 746 } 747 trap_fatal(frame, eva); 748 return (-1); 749 } 750 751 /* kludge to pass faulting virtual address to sendsig */ 752 frame->tf_err = eva; 753 754 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 755} 756#endif 757 758int 759trap_pfault(frame, usermode, eva) 760 struct trapframe *frame; 761 int usermode; 762 vm_offset_t eva; 763{ 764 vm_offset_t va; 765 struct vmspace *vm = NULL; 766 vm_map_t map = 0; 767 int rv = 0; 768 vm_prot_t ftype; 769 struct proc *p = curproc; 770 771 va = trunc_page(eva); 772 if (va >= KERNBASE) { 773 /* 774 * Don't allow user-mode faults in kernel address space. 775 * An exception: if the faulting address is the invalid 776 * instruction entry in the IDT, then the Intel Pentium 777 * F00F bug workaround was triggered, and we need to 778 * treat it is as an illegal instruction, and not a page 779 * fault. 780 */ 781#if defined(I586_CPU) && !defined(NO_F00F_HACK) 782 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) 783 return -2; 784#endif 785 if (usermode) 786 goto nogo; 787 788 map = kernel_map; 789 } else { 790 /* 791 * This is a fault on non-kernel virtual memory. 792 * vm is initialized above to NULL. If curproc is NULL 793 * or curproc->p_vmspace is NULL the fault is fatal. 794 */ 795 if (p != NULL) 796 vm = p->p_vmspace; 797 798 if (vm == NULL) 799 goto nogo; 800 801 map = &vm->vm_map; 802 } 803 804 if (frame->tf_err & PGEX_W) 805 ftype = VM_PROT_WRITE; 806 else 807 ftype = VM_PROT_READ; 808 809 if (map != kernel_map) { 810 /* 811 * Keep swapout from messing with us during this 812 * critical time. 813 */ 814 ++p->p_lock; 815 816 /* 817 * Grow the stack if necessary 818 */ 819 /* grow_stack returns false only if va falls into 820 * a growable stack region and the stack growth 821 * fails. It returns true if va was not within 822 * a growable stack region, or if the stack 823 * growth succeeded. 824 */ 825 if (!grow_stack (p, va)) { 826 rv = KERN_FAILURE; 827 --p->p_lock; 828 goto nogo; 829 } 830 831 /* Fault in the user page: */ 832 rv = vm_fault(map, va, ftype, 833 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY 834 : VM_FAULT_NORMAL); 835 836 --p->p_lock; 837 } else { 838 /* 839 * Don't have to worry about process locking or stacks in the kernel. 840 */ 841 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 842 } 843 844 if (rv == KERN_SUCCESS) 845 return (0); 846nogo: 847 if (!usermode) { 848 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 849 frame->tf_eip = (int)curpcb->pcb_onfault; 850 return (0); 851 } 852 trap_fatal(frame, eva); 853 return (-1); 854 } 855 856 /* kludge to pass faulting virtual address to sendsig */ 857 frame->tf_err = eva; 858 859 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 860} 861 862static void 863trap_fatal(frame, eva) 864 struct trapframe *frame; 865 vm_offset_t eva; 866{ 867 int code, type, ss, esp; 868 struct soft_segment_descriptor softseg; 869 870 code = frame->tf_err; 871 type = frame->tf_trapno; 872 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 873 874 if (type <= MAX_TRAP_MSG) 875 printf("\n\nFatal trap %d: %s while in %s mode\n", 876 type, trap_msg[type], 877 frame->tf_eflags & PSL_VM ? "vm86" : 878 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 879#ifdef SMP 880 /* two seperate prints in case of a trap on an unmapped page */ 881 printf("cpuid = %d; ", cpuid); 882 printf("lapic.id = %08x\n", lapic.id); 883#endif 884 if (type == T_PAGEFLT) { 885 printf("fault virtual address = 0x%x\n", eva); 886 printf("fault code = %s %s, %s\n", 887 code & PGEX_U ? "user" : "supervisor", 888 code & PGEX_W ? "write" : "read", 889 code & PGEX_P ? "protection violation" : "page not present"); 890 } 891 printf("instruction pointer = 0x%x:0x%x\n", 892 frame->tf_cs & 0xffff, frame->tf_eip); 893 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 894 ss = frame->tf_ss & 0xffff; 895 esp = frame->tf_esp; 896 } else { 897 ss = GSEL(GDATA_SEL, SEL_KPL); 898 esp = (int)&frame->tf_esp; 899 } 900 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 901 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 902 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 903 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 904 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 905 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 906 softseg.ssd_gran); 907 printf("processor eflags = "); 908 if (frame->tf_eflags & PSL_T) 909 printf("trace trap, "); 910 if (frame->tf_eflags & PSL_I) 911 printf("interrupt enabled, "); 912 if (frame->tf_eflags & PSL_NT) 913 printf("nested task, "); 914 if (frame->tf_eflags & PSL_RF) 915 printf("resume, "); 916 if (frame->tf_eflags & PSL_VM) 917 printf("vm86, "); 918 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 919 printf("current process = "); 920 if (curproc) { 921 printf("%lu (%s)\n", 922 (u_long)curproc->p_pid, curproc->p_comm ? 923 curproc->p_comm : ""); 924 } else { 925 printf("Idle\n"); 926 } 927 928#ifdef KDB 929 if (kdb_trap(&psl)) 930 return; 931#endif 932#ifdef DDB 933 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame)) 934 return; 935#endif 936 printf("trap number = %d\n", type); 937 if (type <= MAX_TRAP_MSG) 938 panic(trap_msg[type]); 939 else 940 panic("unknown/reserved trap"); 941} 942 943/* 944 * Double fault handler. Called when a fault occurs while writing 945 * a frame for a trap/exception onto the stack. This usually occurs 946 * when the stack overflows (such is the case with infinite recursion, 947 * for example). 948 * 949 * XXX Note that the current PTD gets replaced by IdlePTD when the 950 * task switch occurs. This means that the stack that was active at 951 * the time of the double fault is not available at <kstack> unless 952 * the machine was idle when the double fault occurred. The downside 953 * of this is that "trace <ebp>" in ddb won't work. 954 */ 955void 956dblfault_handler() 957{ 958 printf("\nFatal double fault:\n"); 959 printf("eip = 0x%x\n", common_tss.tss_eip); 960 printf("esp = 0x%x\n", common_tss.tss_esp); 961 printf("ebp = 0x%x\n", common_tss.tss_ebp); 962#ifdef SMP 963 /* two seperate prints in case of a trap on an unmapped page */ 964 printf("cpuid = %d; ", cpuid); 965 printf("lapic.id = %08x\n", lapic.id); 966#endif 967 panic("double fault"); 968} 969 970/* 971 * Compensate for 386 brain damage (missing URKR). 972 * This is a little simpler than the pagefault handler in trap() because 973 * it the page tables have already been faulted in and high addresses 974 * are thrown out early for other reasons. 975 */ 976int trapwrite(addr) 977 unsigned addr; 978{ 979 struct proc *p; 980 vm_offset_t va; 981 struct vmspace *vm; 982 int rv; 983 984 va = trunc_page((vm_offset_t)addr); 985 /* 986 * XXX - MAX is END. Changed > to >= for temp. fix. 987 */ 988 if (va >= VM_MAXUSER_ADDRESS) 989 return (1); 990 991 p = curproc; 992 vm = p->p_vmspace; 993 994 ++p->p_lock; 995 996 if (!grow_stack (p, va)) { 997 --p->p_lock; 998 return (1); 999 } 1000 1001 /* 1002 * fault the data page 1003 */ 1004 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY); 1005 1006 --p->p_lock; 1007 1008 if (rv != KERN_SUCCESS) 1009 return 1; 1010 1011 return (0); 1012} 1013 1014/* 1015 * syscall2 - MP aware system call request C handler 1016 * 1017 * A system call is essentially treated as a trap except that the 1018 * MP lock is not held on entry or return. We are responsible for 1019 * obtaining the MP lock if necessary and for handling ASTs 1020 * (e.g. a task switch) prior to return. 1021 * 1022 * In general, only simple access and manipulation of curproc and 1023 * the current stack is allowed without having to hold MP lock. 1024 */ 1025void 1026syscall2(frame) 1027 struct trapframe frame; 1028{ 1029 caddr_t params; 1030 int i; 1031 struct sysent *callp; 1032 struct proc *p = curproc; 1033 u_quad_t sticks; 1034 int error; 1035 int narg; 1036 int args[8]; 1037 int have_giant = 0; 1038 u_int code; 1039 1040 atomic_add_int(&cnt.v_syscall, 1); 1041 1042#ifdef DIAGNOSTIC 1043 if (ISPL(frame.tf_cs) != SEL_UPL) { 1044 mtx_enter(&Giant, MTX_DEF); 1045 panic("syscall"); 1046 /* NOT REACHED */ 1047 } 1048#endif 1049 1050 /* 1051 * handle atomicy by looping since interrupts are enabled and the 1052 * MP lock is not held. 1053 */ 1054 sticks = ((volatile struct proc *)p)->p_sticks; 1055 while (sticks != ((volatile struct proc *)p)->p_sticks) 1056 sticks = ((volatile struct proc *)p)->p_sticks; 1057 1058 p->p_md.md_regs = &frame; 1059 params = (caddr_t)frame.tf_esp + sizeof(int); 1060 code = frame.tf_eax; 1061 1062 if (p->p_sysent->sv_prepsyscall) { 1063 /* 1064 * The prep code is not MP aware. 1065 */ 1066 mtx_enter(&Giant, MTX_DEF); 1067 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 1068 mtx_exit(&Giant, MTX_DEF); 1069 } else { 1070 /* 1071 * Need to check if this is a 32 bit or 64 bit syscall. 1072 * fuword is MP aware. 1073 */ 1074 if (code == SYS_syscall) { 1075 /* 1076 * Code is first argument, followed by actual args. 1077 */ 1078 code = fuword(params); 1079 params += sizeof(int); 1080 } else if (code == SYS___syscall) { 1081 /* 1082 * Like syscall, but code is a quad, so as to maintain 1083 * quad alignment for the rest of the arguments. 1084 */ 1085 code = fuword(params); 1086 params += sizeof(quad_t); 1087 } 1088 } 1089 1090 if (p->p_sysent->sv_mask) 1091 code &= p->p_sysent->sv_mask; 1092 1093 if (code >= p->p_sysent->sv_size) 1094 callp = &p->p_sysent->sv_table[0]; 1095 else 1096 callp = &p->p_sysent->sv_table[code]; 1097 1098 narg = callp->sy_narg & SYF_ARGMASK; 1099 1100 /* 1101 * copyin is MP aware, but the tracing code is not 1102 */ 1103 if (params && (i = narg * sizeof(int)) && 1104 (error = copyin(params, (caddr_t)args, (u_int)i))) { 1105 mtx_enter(&Giant, MTX_DEF); 1106 have_giant = 1; 1107#ifdef KTRACE 1108 if (KTRPOINT(p, KTR_SYSCALL)) 1109 ktrsyscall(p->p_tracep, code, narg, args); 1110#endif 1111 goto bad; 1112 } 1113 1114 /* 1115 * Try to run the syscall without the MP lock if the syscall 1116 * is MP safe. We have to obtain the MP lock no matter what if 1117 * we are ktracing 1118 */ 1119 if ((callp->sy_narg & SYF_MPSAFE) == 0) { 1120 mtx_enter(&Giant, MTX_DEF); 1121 have_giant = 1; 1122 } 1123 1124#ifdef KTRACE 1125 if (KTRPOINT(p, KTR_SYSCALL)) { 1126 if (have_giant == 0) { 1127 mtx_enter(&Giant, MTX_DEF); 1128 have_giant = 1; 1129 } 1130 ktrsyscall(p->p_tracep, code, narg, args); 1131 } 1132#endif 1133 p->p_retval[0] = 0; 1134 p->p_retval[1] = frame.tf_edx; 1135 1136 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1137 1138 error = (*callp->sy_call)(p, args); 1139 1140 /* 1141 * MP SAFE (we may or may not have the MP lock at this point) 1142 */ 1143 switch (error) { 1144 case 0: 1145 /* 1146 * Reinitialize proc pointer `p' as it may be different 1147 * if this is a child returning from fork syscall. 1148 */ 1149 p = curproc; 1150 frame.tf_eax = p->p_retval[0]; 1151 frame.tf_edx = p->p_retval[1]; 1152 frame.tf_eflags &= ~PSL_C; 1153 break; 1154 1155 case ERESTART: 1156 /* 1157 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1158 * int 0x80 is 2 bytes. We saved this in tf_err. 1159 */ 1160 frame.tf_eip -= frame.tf_err; 1161 break; 1162 1163 case EJUSTRETURN: 1164 break; 1165 1166 default: 1167bad: 1168 if (p->p_sysent->sv_errsize) { 1169 if (error >= p->p_sysent->sv_errsize) 1170 error = -1; /* XXX */ 1171 else 1172 error = p->p_sysent->sv_errtbl[error]; 1173 } 1174 frame.tf_eax = error; 1175 frame.tf_eflags |= PSL_C; 1176 break; 1177 } 1178 1179 /* 1180 * Traced syscall. trapsignal() is not MP aware. 1181 */ 1182 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1183 if (have_giant == 0) { 1184 mtx_enter(&Giant, MTX_DEF); 1185 have_giant = 1; 1186 } 1187 frame.tf_eflags &= ~PSL_T; 1188 trapsignal(p, SIGTRAP, 0); 1189 } 1190 1191 /* 1192 * Handle reschedule and other end-of-syscall issues 1193 */ 1194 have_giant = userret(p, &frame, sticks, have_giant); 1195 1196#ifdef KTRACE 1197 if (KTRPOINT(p, KTR_SYSRET)) { 1198 if (have_giant == 0) { 1199 mtx_enter(&Giant, MTX_DEF); 1200 have_giant = 1; 1201 } 1202 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1203 } 1204#endif 1205 1206 /* 1207 * This works because errno is findable through the 1208 * register set. If we ever support an emulation where this 1209 * is not the case, this code will need to be revisited. 1210 */ 1211 STOPEVENT(p, S_SCX, code); 1212 1213 /* 1214 * Release the MP lock if we had to get it 1215 */ 1216 if (have_giant) 1217 mtx_exit(&Giant, MTX_DEF); 1218 1219 mtx_assert(&sched_lock, MA_NOTOWNED); 1220 mtx_assert(&Giant, MA_NOTOWNED); 1221} 1222 1223void 1224ast(frame) 1225 struct trapframe frame; 1226{ 1227 struct proc *p = CURPROC; 1228 u_quad_t sticks; 1229 1230 /* 1231 * handle atomicy by looping since interrupts are enabled and the 1232 * MP lock is not held. 1233 */ 1234 sticks = ((volatile struct proc *)p)->p_sticks; 1235 while (sticks != ((volatile struct proc *)p)->p_sticks) 1236 sticks = ((volatile struct proc *)p)->p_sticks; 1237 1238 astoff(); 1239 atomic_add_int(&cnt.v_soft, 1); 1240 if (p->p_flag & P_OWEUPC) { 1241 mtx_enter(&Giant, MTX_DEF); 1242 p->p_flag &= ~P_OWEUPC; 1243 addupc_task(p, p->p_stats->p_prof.pr_addr, 1244 p->p_stats->p_prof.pr_ticks); 1245} 1246 if (userret(p, &frame, sticks, mtx_owned(&Giant)) != 0) 1247 mtx_exit(&Giant, MTX_DEF); 1248} 1249 1250/* 1251 * Simplified back end of syscall(), used when returning from fork() 1252 * directly into user mode. Giant is not held on entry, and must not 1253 * be held on return. 1254 */ 1255void 1256fork_return(p, frame) 1257 struct proc *p; 1258 struct trapframe frame; 1259{ 1260 int have_giant; 1261 1262 frame.tf_eax = 0; /* Child returns zero */ 1263 frame.tf_eflags &= ~PSL_C; /* success */ 1264 frame.tf_edx = 1; 1265 1266 have_giant = userret(p, &frame, 0, mtx_owned(&Giant)); 1267#ifdef KTRACE 1268 if (KTRPOINT(p, KTR_SYSRET)) { 1269 if (have_giant == 0) { 1270 mtx_enter(&Giant, MTX_DEF); 1271 have_giant = 1; 1272 } 1273 ktrsysret(p->p_tracep, SYS_fork, 0, 0); 1274 } 1275#endif 1276 if (have_giant) 1277 mtx_exit(&Giant, MTX_DEF); 1278} 1279