subr_syscall.c revision 33134
1208963Srdivacky/*- 2208963Srdivacky * Copyright (C) 1994, David Greenman 3208963Srdivacky * Copyright (c) 1990, 1993 4208963Srdivacky * The Regents of the University of California. All rights reserved. 5208963Srdivacky * 6208963Srdivacky * This code is derived from software contributed to Berkeley by 7208963Srdivacky * the University of Utah, and William Jolitz. 8208963Srdivacky * 9208963Srdivacky * Redistribution and use in source and binary forms, with or without 10208963Srdivacky * modification, are permitted provided that the following conditions 11208963Srdivacky * are met: 12208963Srdivacky * 1. Redistributions of source code must retain the above copyright 13208963Srdivacky * notice, this list of conditions and the following disclaimer. 14208963Srdivacky * 2. Redistributions in binary form must reproduce the above copyright 15208963Srdivacky * notice, this list of conditions and the following disclaimer in the 16208963Srdivacky * documentation and/or other materials provided with the distribution. 17208963Srdivacky * 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 * $Id: trap.c,v 1.121 1998/02/04 22:32:12 eivind Exp $ 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_trap.h" 49#include "opt_vm86.h" 50 51#include <sys/param.h> 52#include <sys/systm.h> 53#include <sys/proc.h> 54#include <sys/pioctl.h> 55#include <sys/kernel.h> 56#include <sys/resourcevar.h> 57#include <sys/signalvar.h> 58#include <sys/syscall.h> 59#include <sys/sysent.h> 60#include <sys/vmmeter.h> 61#ifdef KTRACE 62#include <sys/ktrace.h> 63#endif 64 65#include <vm/vm.h> 66#include <vm/vm_param.h> 67#include <vm/vm_prot.h> 68#include <sys/lock.h> 69#include <vm/pmap.h> 70#include <vm/vm_kern.h> 71#include <vm/vm_map.h> 72#include <vm/vm_page.h> 73#include <vm/vm_extern.h> 74 75#include <machine/cpu.h> 76#include <machine/ipl.h> 77#include <machine/md_var.h> 78#include <machine/pcb.h> 79#ifdef SMP 80#include <machine/smp.h> 81#endif 82#include <machine/tss.h> 83 84#include <i386/isa/intr_machdep.h> 85 86#ifdef POWERFAIL_NMI 87#include <sys/syslog.h> 88#include <machine/clock.h> 89#endif 90 91#ifdef VM86 92#include <machine/vm86.h> 93#endif 94 95#include "isa.h" 96#include "npx.h" 97 98extern struct i386tss common_tss; 99 100int (*pmath_emulate) __P((struct trapframe *)); 101 102extern void trap __P((struct trapframe frame)); 103extern int trapwrite __P((unsigned addr)); 104extern void syscall __P((struct trapframe frame)); 105 106static int trap_pfault __P((struct trapframe *, int)); 107static void trap_fatal __P((struct trapframe *)); 108void dblfault_handler __P((void)); 109 110extern inthand_t IDTVEC(syscall); 111 112#define MAX_TRAP_MSG 28 113static char *trap_msg[] = { 114 "", /* 0 unused */ 115 "privileged instruction fault", /* 1 T_PRIVINFLT */ 116 "", /* 2 unused */ 117 "breakpoint instruction fault", /* 3 T_BPTFLT */ 118 "", /* 4 unused */ 119 "", /* 5 unused */ 120 "arithmetic trap", /* 6 T_ARITHTRAP */ 121 "system forced exception", /* 7 T_ASTFLT */ 122 "", /* 8 unused */ 123 "general protection fault", /* 9 T_PROTFLT */ 124 "trace trap", /* 10 T_TRCTRAP */ 125 "", /* 11 unused */ 126 "page fault", /* 12 T_PAGEFLT */ 127 "", /* 13 unused */ 128 "alignment fault", /* 14 T_ALIGNFLT */ 129 "", /* 15 unused */ 130 "", /* 16 unused */ 131 "", /* 17 unused */ 132 "integer divide fault", /* 18 T_DIVIDE */ 133 "non-maskable interrupt trap", /* 19 T_NMI */ 134 "overflow trap", /* 20 T_OFLOW */ 135 "FPU bounds check fault", /* 21 T_BOUND */ 136 "FPU device not available", /* 22 T_DNA */ 137 "double fault", /* 23 T_DOUBLEFLT */ 138 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 139 "invalid TSS fault", /* 25 T_TSSFLT */ 140 "segment not present fault", /* 26 T_SEGNPFLT */ 141 "stack fault", /* 27 T_STKFLT */ 142 "machine check trap", /* 28 T_MCHK */ 143}; 144 145static void userret __P((struct proc *p, struct trapframe *frame, 146 u_quad_t oticks)); 147 148#if defined(I586_CPU) && !defined(NO_F00F_HACK) 149extern struct gate_descriptor *t_idt; 150extern int has_f00f_bug; 151#endif 152 153static inline void 154userret(p, frame, oticks) 155 struct proc *p; 156 struct trapframe *frame; 157 u_quad_t oticks; 158{ 159 int sig, s; 160 161 while ((sig = CURSIG(p)) != 0) 162 postsig(sig); 163 164#if 0 165 if (!want_resched && 166 (p->p_priority <= p->p_usrpri) && 167 (p->p_rtprio.type == RTP_PRIO_NORMAL)) { 168 int newpriority; 169 p->p_estcpu += 1; 170 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice; 171 newpriority = min(newpriority, MAXPRI); 172 p->p_usrpri = newpriority; 173 } 174#endif 175 176 p->p_priority = p->p_usrpri; 177 if (want_resched) { 178 /* 179 * Since we are curproc, clock will normally just change 180 * our priority without moving us from one queue to another 181 * (since the running process is not on a queue.) 182 * If that happened after we setrunqueue ourselves but before we 183 * mi_switch()'ed, we might not be on the queue indicated by 184 * our priority. 185 */ 186 s = splhigh(); 187 setrunqueue(p); 188 p->p_stats->p_ru.ru_nivcsw++; 189 mi_switch(); 190 splx(s); 191 while ((sig = CURSIG(p)) != 0) 192 postsig(sig); 193 } 194 /* 195 * Charge system time if profiling. 196 */ 197 if (p->p_flag & P_PROFIL) 198 addupc_task(p, frame->tf_eip, 199 (u_int)(p->p_sticks - oticks) * psratio); 200 201 curpriority = p->p_priority; 202} 203 204/* 205 * Exception, fault, and trap interface to the FreeBSD kernel. 206 * This common code is called from assembly language IDT gate entry 207 * routines that prepare a suitable stack frame, and restore this 208 * frame after the exception has been processed. 209 */ 210 211void 212trap(frame) 213 struct trapframe frame; 214{ 215 struct proc *p = curproc; 216 u_quad_t sticks = 0; 217 int i = 0, ucode = 0, type, code; 218#ifdef DEBUG 219 u_long eva; 220#endif 221 222#if defined(I586_CPU) && !defined(NO_F00F_HACK) 223restart: 224#endif 225 type = frame.tf_trapno; 226 code = frame.tf_err; 227 228 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) { 229 /* user trap */ 230 231 sticks = p->p_sticks; 232 p->p_md.md_regs = &frame; 233 234 switch (type) { 235 case T_PRIVINFLT: /* privileged instruction fault */ 236 ucode = type; 237 i = SIGILL; 238 break; 239 240 case T_BPTFLT: /* bpt instruction fault */ 241 case T_TRCTRAP: /* trace trap */ 242 frame.tf_eflags &= ~PSL_T; 243 i = SIGTRAP; 244 break; 245 246 case T_ARITHTRAP: /* arithmetic trap */ 247 ucode = code; 248 i = SIGFPE; 249 break; 250 251 case T_ASTFLT: /* Allow process switch */ 252 astoff(); 253 cnt.v_soft++; 254 if (p->p_flag & P_OWEUPC) { 255 p->p_flag &= ~P_OWEUPC; 256 addupc_task(p, p->p_stats->p_prof.pr_addr, 257 p->p_stats->p_prof.pr_ticks); 258 } 259 goto out; 260 261 /* 262 * The following two traps can happen in 263 * vm86 mode, and, if so, we want to handle 264 * them specially. 265 */ 266 case T_PROTFLT: /* general protection fault */ 267 case T_STKFLT: /* stack fault */ 268#ifdef VM86 269 if (frame.tf_eflags & PSL_VM) { 270 i = vm86_emulate((struct vm86frame *)&frame); 271 if (i == 0) 272 goto out; 273 break; 274 } 275#endif /* VM86 */ 276 /* FALL THROUGH */ 277 278 case T_SEGNPFLT: /* segment not present fault */ 279 case T_TSSFLT: /* invalid TSS fault */ 280 case T_DOUBLEFLT: /* double fault */ 281 default: 282 ucode = code + BUS_SEGM_FAULT ; 283 i = SIGBUS; 284 break; 285 286 case T_PAGEFLT: /* page fault */ 287 i = trap_pfault(&frame, TRUE); 288 if (i == -1) 289 return; 290#if defined(I586_CPU) && !defined(NO_F00F_HACK) 291 if (i == -2) 292 goto restart; 293#endif 294 if (i == 0) 295 goto out; 296 297 ucode = T_PAGEFLT; 298 break; 299 300 case T_DIVIDE: /* integer divide fault */ 301 ucode = FPE_INTDIV_TRAP; 302 i = SIGFPE; 303 break; 304 305#if NISA > 0 306 case T_NMI: 307#ifdef POWERFAIL_NMI 308 goto handle_powerfail; 309#else /* !POWERFAIL_NMI */ 310#ifdef DDB 311 /* NMI can be hooked up to a pushbutton for debugging */ 312 printf ("NMI ... going to debugger\n"); 313 if (kdb_trap (type, 0, &frame)) 314 return; 315#endif /* DDB */ 316 /* machine/parity/power fail/"kitchen sink" faults */ 317 if (isa_nmi(code) == 0) return; 318 panic("NMI indicates hardware failure"); 319#endif /* POWERFAIL_NMI */ 320#endif /* NISA > 0 */ 321 322 case T_OFLOW: /* integer overflow fault */ 323 ucode = FPE_INTOVF_TRAP; 324 i = SIGFPE; 325 break; 326 327 case T_BOUND: /* bounds check fault */ 328 ucode = FPE_SUBRNG_TRAP; 329 i = SIGFPE; 330 break; 331 332 case T_DNA: 333#if NNPX > 0 334 /* if a transparent fault (due to context switch "late") */ 335 if (npxdna()) 336 return; 337#endif 338 if (!pmath_emulate) { 339 i = SIGFPE; 340 ucode = FPE_FPU_NP_TRAP; 341 break; 342 } 343 i = (*pmath_emulate)(&frame); 344 if (i == 0) { 345 if (!(frame.tf_eflags & PSL_T)) 346 return; 347 frame.tf_eflags &= ~PSL_T; 348 i = SIGTRAP; 349 } 350 /* else ucode = emulator_only_knows() XXX */ 351 break; 352 353 case T_FPOPFLT: /* FPU operand fetch fault */ 354 ucode = T_FPOPFLT; 355 i = SIGILL; 356 break; 357 } 358 } else { 359 /* kernel trap */ 360 361 switch (type) { 362 case T_PAGEFLT: /* page fault */ 363 (void) trap_pfault(&frame, FALSE); 364 return; 365 366 case T_DNA: 367#if NNPX > 0 368 /* 369 * The kernel is apparently using npx for copying. 370 * XXX this should be fatal unless the kernel has 371 * registered such use. 372 */ 373 if (npxdna()) 374 return; 375#endif 376 break; 377 378 case T_PROTFLT: /* general protection fault */ 379 case T_SEGNPFLT: /* segment not present fault */ 380 /* 381 * Invalid segment selectors and out of bounds 382 * %eip's and %esp's can be set up in user mode. 383 * This causes a fault in kernel mode when the 384 * kernel tries to return to user mode. We want 385 * to get this fault so that we can fix the 386 * problem here and not have to check all the 387 * selectors and pointers when the user changes 388 * them. 389 */ 390#define MAYBE_DORETI_FAULT(where, whereto) \ 391 do { \ 392 if (frame.tf_eip == (int)where) { \ 393 frame.tf_eip = (int)whereto; \ 394 return; \ 395 } \ 396 } while (0) 397 398 if (intr_nesting_level == 0) { 399 /* 400 * Invalid %fs's and %gs's can be created using 401 * procfs or PT_SETREGS or by invalidating the 402 * underlying LDT entry. This causes a fault 403 * in kernel mode when the kernel attempts to 404 * switch contexts. Lose the bad context 405 * (XXX) so that we can continue, and generate 406 * a signal. 407 */ 408 if (frame.tf_eip == (int)cpu_switch_load_fs) { 409 curpcb->pcb_fs = 0; 410 psignal(p, SIGBUS); 411 return; 412 } 413 if (frame.tf_eip == (int)cpu_switch_load_gs) { 414 curpcb->pcb_gs = 0; 415 psignal(p, SIGBUS); 416 return; 417 } 418 MAYBE_DORETI_FAULT(doreti_iret, 419 doreti_iret_fault); 420 MAYBE_DORETI_FAULT(doreti_popl_ds, 421 doreti_popl_ds_fault); 422 MAYBE_DORETI_FAULT(doreti_popl_es, 423 doreti_popl_es_fault); 424 if (curpcb && curpcb->pcb_onfault) { 425 frame.tf_eip = (int)curpcb->pcb_onfault; 426 return; 427 } 428 } 429 break; 430 431 case T_TSSFLT: 432 /* 433 * PSL_NT can be set in user mode and isn't cleared 434 * automatically when the kernel is entered. This 435 * causes a TSS fault when the kernel attempts to 436 * `iret' because the TSS link is uninitialized. We 437 * want to get this fault so that we can fix the 438 * problem here and not every time the kernel is 439 * entered. 440 */ 441 if (frame.tf_eflags & PSL_NT) { 442 frame.tf_eflags &= ~PSL_NT; 443 return; 444 } 445 break; 446 447 case T_TRCTRAP: /* trace trap */ 448 if (frame.tf_eip == (int)IDTVEC(syscall)) { 449 /* 450 * We've just entered system mode via the 451 * syscall lcall. Continue single stepping 452 * silently until the syscall handler has 453 * saved the flags. 454 */ 455 return; 456 } 457 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { 458 /* 459 * The syscall handler has now saved the 460 * flags. Stop single stepping it. 461 */ 462 frame.tf_eflags &= ~PSL_T; 463 return; 464 } 465 /* 466 * Fall through. 467 */ 468 case T_BPTFLT: 469 /* 470 * If DDB is enabled, let it handle the debugger trap. 471 * Otherwise, debugger traps "can't happen". 472 */ 473#ifdef DDB 474 if (kdb_trap (type, 0, &frame)) 475 return; 476#endif 477 break; 478 479#if NISA > 0 480 case T_NMI: 481#ifdef POWERFAIL_NMI 482#ifndef TIMER_FREQ 483# define TIMER_FREQ 1193182 484#endif 485 handle_powerfail: 486 { 487 static unsigned lastalert = 0; 488 489 if(time.tv_sec - lastalert > 10) 490 { 491 log(LOG_WARNING, "NMI: power fail\n"); 492 sysbeep(TIMER_FREQ/880, hz); 493 lastalert = time.tv_sec; 494 } 495 return; 496 } 497#else /* !POWERFAIL_NMI */ 498#ifdef DDB 499 /* NMI can be hooked up to a pushbutton for debugging */ 500 printf ("NMI ... going to debugger\n"); 501 if (kdb_trap (type, 0, &frame)) 502 return; 503#endif /* DDB */ 504 /* machine/parity/power fail/"kitchen sink" faults */ 505 if (isa_nmi(code) == 0) return; 506 /* FALL THROUGH */ 507#endif /* POWERFAIL_NMI */ 508#endif /* NISA > 0 */ 509 } 510 511 trap_fatal(&frame); 512 return; 513 } 514 515 trapsignal(p, i, ucode); 516 517#ifdef DEBUG 518 eva = rcr2(); 519 if (type <= MAX_TRAP_MSG) { 520 uprintf("fatal process exception: %s", 521 trap_msg[type]); 522 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 523 uprintf(", fault VA = 0x%x", eva); 524 uprintf("\n"); 525 } 526#endif 527 528out: 529 userret(p, &frame, sticks); 530} 531 532#ifdef notyet 533/* 534 * This version doesn't allow a page fault to user space while 535 * in the kernel. The rest of the kernel needs to be made "safe" 536 * before this can be used. I think the only things remaining 537 * to be made safe are the iBCS2 code and the process tracing/ 538 * debugging code. 539 */ 540static int 541trap_pfault(frame, usermode) 542 struct trapframe *frame; 543 int usermode; 544{ 545 vm_offset_t va; 546 struct vmspace *vm = NULL; 547 vm_map_t map = 0; 548 int rv = 0; 549 vm_prot_t ftype; 550 int eva; 551 struct proc *p = curproc; 552 553 if (frame->tf_err & PGEX_W) 554 ftype = VM_PROT_READ | VM_PROT_WRITE; 555 else 556 ftype = VM_PROT_READ; 557 558 eva = rcr2(); 559 va = trunc_page((vm_offset_t)eva); 560 561 if (va < VM_MIN_KERNEL_ADDRESS) { 562 vm_offset_t v; 563 vm_page_t mpte; 564 565 if (p == NULL || 566 (!usermode && va < VM_MAXUSER_ADDRESS && 567 (intr_nesting_level != 0 || curpcb == NULL || 568 curpcb->pcb_onfault == NULL))) { 569 trap_fatal(frame); 570 return (-1); 571 } 572 573 /* 574 * This is a fault on non-kernel virtual memory. 575 * vm is initialized above to NULL. If curproc is NULL 576 * or curproc->p_vmspace is NULL the fault is fatal. 577 */ 578 vm = p->p_vmspace; 579 if (vm == NULL) 580 goto nogo; 581 582 map = &vm->vm_map; 583 584 /* 585 * Keep swapout from messing with us during this 586 * critical time. 587 */ 588 ++p->p_lock; 589 590 /* 591 * Grow the stack if necessary 592 */ 593 if ((caddr_t)va > vm->vm_maxsaddr 594 && (caddr_t)va < (caddr_t)USRSTACK) { 595 if (!grow(p, va)) { 596 rv = KERN_FAILURE; 597 --p->p_lock; 598 goto nogo; 599 } 600 } 601 602 /* Fault in the user page: */ 603 rv = vm_fault(map, va, ftype, 604 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0); 605 606 --p->p_lock; 607 } else { 608 /* 609 * Don't allow user-mode faults in kernel address space. 610 */ 611 if (usermode) 612 goto nogo; 613 614 /* 615 * Since we know that kernel virtual address addresses 616 * always have pte pages mapped, we just have to fault 617 * the page. 618 */ 619 rv = vm_fault(kernel_map, va, ftype, FALSE); 620 } 621 622 if (rv == KERN_SUCCESS) 623 return (0); 624nogo: 625 if (!usermode) { 626 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 627 frame->tf_eip = (int)curpcb->pcb_onfault; 628 return (0); 629 } 630 trap_fatal(frame); 631 return (-1); 632 } 633 634 /* kludge to pass faulting virtual address to sendsig */ 635 frame->tf_err = eva; 636 637 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 638} 639#endif 640 641int 642trap_pfault(frame, usermode) 643 struct trapframe *frame; 644 int usermode; 645{ 646 vm_offset_t va; 647 struct vmspace *vm = NULL; 648 vm_map_t map = 0; 649 int rv = 0; 650 vm_prot_t ftype; 651 int eva; 652 struct proc *p = curproc; 653 654 eva = rcr2(); 655 va = trunc_page((vm_offset_t)eva); 656 657 if (va >= KERNBASE) { 658 /* 659 * Don't allow user-mode faults in kernel address space. 660 * An exception: if the faulting address is the invalid 661 * instruction entry in the IDT, then the Intel Pentium 662 * F00F bug workaround was triggered, and we need to 663 * treat it is as an illegal instruction, and not a page 664 * fault. 665 */ 666#if defined(I586_CPU) && !defined(NO_F00F_HACK) 667 if ((eva == (unsigned int)&t_idt[6]) && has_f00f_bug) { 668 frame->tf_trapno = T_PRIVINFLT; 669 return -2; 670 } 671#endif 672 if (usermode) 673 goto nogo; 674 675 map = kernel_map; 676 } else { 677 /* 678 * This is a fault on non-kernel virtual memory. 679 * vm is initialized above to NULL. If curproc is NULL 680 * or curproc->p_vmspace is NULL the fault is fatal. 681 */ 682 if (p != NULL) 683 vm = p->p_vmspace; 684 685 if (vm == NULL) 686 goto nogo; 687 688 map = &vm->vm_map; 689 } 690 691 if (frame->tf_err & PGEX_W) 692 ftype = VM_PROT_READ | VM_PROT_WRITE; 693 else 694 ftype = VM_PROT_READ; 695 696 if (map != kernel_map) { 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 if ((caddr_t)va > vm->vm_maxsaddr 707 && (caddr_t)va < (caddr_t)USRSTACK) { 708 if (!grow(p, va)) { 709 rv = KERN_FAILURE; 710 --p->p_lock; 711 goto nogo; 712 } 713 } 714 715 /* Fault in the user page: */ 716 rv = vm_fault(map, va, ftype, 717 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0); 718 719 --p->p_lock; 720 } else { 721 /* 722 * Don't have to worry about process locking or stacks in the kernel. 723 */ 724 rv = vm_fault(map, va, ftype, FALSE); 725 } 726 727 if (rv == KERN_SUCCESS) 728 return (0); 729nogo: 730 if (!usermode) { 731 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 732 frame->tf_eip = (int)curpcb->pcb_onfault; 733 return (0); 734 } 735 trap_fatal(frame); 736 return (-1); 737 } 738 739 /* kludge to pass faulting virtual address to sendsig */ 740 frame->tf_err = eva; 741 742 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 743} 744 745static void 746trap_fatal(frame) 747 struct trapframe *frame; 748{ 749 int code, type, eva, ss, esp; 750 struct soft_segment_descriptor softseg; 751 752 code = frame->tf_err; 753 type = frame->tf_trapno; 754 eva = rcr2(); 755 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 756 757 if (type <= MAX_TRAP_MSG) 758 printf("\n\nFatal trap %d: %s while in %s mode\n", 759 type, trap_msg[type], 760 frame->tf_eflags & PSL_VM ? "vm86" : 761 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 762#ifdef SMP 763 /* three seperate prints in case of a trap on an unmapped page */ 764 printf("mp_lock = %08x; ", mp_lock); 765 printf("cpuid = %d; ", cpuid); 766 printf("lapic.id = %08x\n", lapic.id); 767#endif 768 if (type == T_PAGEFLT) { 769 printf("fault virtual address = 0x%x\n", eva); 770 printf("fault code = %s %s, %s\n", 771 code & PGEX_U ? "user" : "supervisor", 772 code & PGEX_W ? "write" : "read", 773 code & PGEX_P ? "protection violation" : "page not present"); 774 } 775 printf("instruction pointer = 0x%x:0x%x\n", 776 frame->tf_cs & 0xffff, frame->tf_eip); 777 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 778 ss = frame->tf_ss & 0xffff; 779 esp = frame->tf_esp; 780 } else { 781 ss = GSEL(GDATA_SEL, SEL_KPL); 782 esp = (int)&frame->tf_esp; 783 } 784 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 785 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 786 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 787 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 788 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 789 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 790 softseg.ssd_gran); 791 printf("processor eflags = "); 792 if (frame->tf_eflags & PSL_T) 793 printf("trace trap, "); 794 if (frame->tf_eflags & PSL_I) 795 printf("interrupt enabled, "); 796 if (frame->tf_eflags & PSL_NT) 797 printf("nested task, "); 798 if (frame->tf_eflags & PSL_RF) 799 printf("resume, "); 800 if (frame->tf_eflags & PSL_VM) 801 printf("vm86, "); 802 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 803 printf("current process = "); 804 if (curproc) { 805 printf("%lu (%s)\n", 806 (u_long)curproc->p_pid, curproc->p_comm ? 807 curproc->p_comm : ""); 808 } else { 809 printf("Idle\n"); 810 } 811 printf("interrupt mask = "); 812 if ((cpl & net_imask) == net_imask) 813 printf("net "); 814 if ((cpl & tty_imask) == tty_imask) 815 printf("tty "); 816 if ((cpl & bio_imask) == bio_imask) 817 printf("bio "); 818 if ((cpl & cam_imask) == cam_imask) 819 printf("cam "); 820 if (cpl == 0) 821 printf("none"); 822#ifdef SMP 823/** 824 * XXX FIXME: 825 * we probably SHOULD have stopped the other CPUs before now! 826 * another CPU COULD have been touching cpl at this moment... 827 */ 828 printf(" <- SMP: XXX"); 829#endif 830 printf("\n"); 831 832#ifdef KDB 833 if (kdb_trap(&psl)) 834 return; 835#endif 836#ifdef DDB 837 if (kdb_trap (type, 0, frame)) 838 return; 839#endif 840 printf("trap number = %d\n", type); 841 if (type <= MAX_TRAP_MSG) 842 panic(trap_msg[type]); 843 else 844 panic("unknown/reserved trap"); 845} 846 847/* 848 * Double fault handler. Called when a fault occurs while writing 849 * a frame for a trap/exception onto the stack. This usually occurs 850 * when the stack overflows (such is the case with infinite recursion, 851 * for example). 852 * 853 * XXX Note that the current PTD gets replaced by IdlePTD when the 854 * task switch occurs. This means that the stack that was active at 855 * the time of the double fault is not available at <kstack> unless 856 * the machine was idle when the double fault occurred. The downside 857 * of this is that "trace <ebp>" in ddb won't work. 858 */ 859void 860dblfault_handler() 861{ 862 printf("\nFatal double fault:\n"); 863 printf("eip = 0x%x\n", common_tss.tss_eip); 864 printf("esp = 0x%x\n", common_tss.tss_esp); 865 printf("ebp = 0x%x\n", common_tss.tss_ebp); 866#ifdef SMP 867 /* three seperate prints in case of a trap on an unmapped page */ 868 printf("mp_lock = %08x; ", mp_lock); 869 printf("cpuid = %d; ", cpuid); 870 printf("lapic.id = %08x\n", lapic.id); 871#endif 872 panic("double fault"); 873} 874 875/* 876 * Compensate for 386 brain damage (missing URKR). 877 * This is a little simpler than the pagefault handler in trap() because 878 * it the page tables have already been faulted in and high addresses 879 * are thrown out early for other reasons. 880 */ 881int trapwrite(addr) 882 unsigned addr; 883{ 884 struct proc *p; 885 vm_offset_t va; 886 struct vmspace *vm; 887 int rv; 888 889 va = trunc_page((vm_offset_t)addr); 890 /* 891 * XXX - MAX is END. Changed > to >= for temp. fix. 892 */ 893 if (va >= VM_MAXUSER_ADDRESS) 894 return (1); 895 896 p = curproc; 897 vm = p->p_vmspace; 898 899 ++p->p_lock; 900 901 if ((caddr_t)va >= vm->vm_maxsaddr 902 && (caddr_t)va < (caddr_t)USRSTACK) { 903 if (!grow(p, va)) { 904 --p->p_lock; 905 return (1); 906 } 907 } 908 909 /* 910 * fault the data page 911 */ 912 rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY); 913 914 --p->p_lock; 915 916 if (rv != KERN_SUCCESS) 917 return 1; 918 919 return (0); 920} 921 922/* 923 * System call request from POSIX system call gate interface to kernel. 924 * Like trap(), argument is call by reference. 925 */ 926void 927syscall(frame) 928 struct trapframe frame; 929{ 930 caddr_t params; 931 int i; 932 struct sysent *callp; 933 struct proc *p = curproc; 934 u_quad_t sticks; 935 int error; 936 int args[8]; 937 u_int code; 938 939#ifdef DIAGNOSTIC 940 if (ISPL(frame.tf_cs) != SEL_UPL) 941 panic("syscall"); 942#endif 943 sticks = p->p_sticks; 944 p->p_md.md_regs = &frame; 945 params = (caddr_t)frame.tf_esp + sizeof(int); 946 code = frame.tf_eax; 947 if (p->p_sysent->sv_prepsyscall) { 948 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 949 } else { 950 /* 951 * Need to check if this is a 32 bit or 64 bit syscall. 952 */ 953 if (code == SYS_syscall) { 954 /* 955 * Code is first argument, followed by actual args. 956 */ 957 code = fuword(params); 958 params += sizeof(int); 959 } else if (code == SYS___syscall) { 960 /* 961 * Like syscall, but code is a quad, so as to maintain 962 * quad alignment for the rest of the arguments. 963 */ 964 code = fuword(params); 965 params += sizeof(quad_t); 966 } 967 } 968 969 if (p->p_sysent->sv_mask) 970 code &= p->p_sysent->sv_mask; 971 972 if (code >= p->p_sysent->sv_size) 973 callp = &p->p_sysent->sv_table[0]; 974 else 975 callp = &p->p_sysent->sv_table[code]; 976 977 if (params && (i = callp->sy_narg * sizeof(int)) && 978 (error = copyin(params, (caddr_t)args, (u_int)i))) { 979#ifdef KTRACE 980 if (KTRPOINT(p, KTR_SYSCALL)) 981 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 982#endif 983 goto bad; 984 } 985#ifdef KTRACE 986 if (KTRPOINT(p, KTR_SYSCALL)) 987 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 988#endif 989 p->p_retval[0] = 0; 990 p->p_retval[1] = frame.tf_edx; 991 992 STOPEVENT(p, S_SCE, callp->sy_narg); 993 994 error = (*callp->sy_call)(p, args); 995 996 switch (error) { 997 998 case 0: 999 /* 1000 * Reinitialize proc pointer `p' as it may be different 1001 * if this is a child returning from fork syscall. 1002 */ 1003 p = curproc; 1004 frame.tf_eax = p->p_retval[0]; 1005 frame.tf_edx = p->p_retval[1]; 1006 frame.tf_eflags &= ~PSL_C; 1007 break; 1008 1009 case ERESTART: 1010 /* 1011 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1012 * int 0x80 is 2 bytes. We saved this in tf_err. 1013 */ 1014 frame.tf_eip -= frame.tf_err; 1015 break; 1016 1017 case EJUSTRETURN: 1018 break; 1019 1020 default: 1021bad: 1022 if (p->p_sysent->sv_errsize) 1023 if (error >= p->p_sysent->sv_errsize) 1024 error = -1; /* XXX */ 1025 else 1026 error = p->p_sysent->sv_errtbl[error]; 1027 frame.tf_eax = error; 1028 frame.tf_eflags |= PSL_C; 1029 break; 1030 } 1031 1032 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1033 /* Traced syscall. */ 1034 frame.tf_eflags &= ~PSL_T; 1035 trapsignal(p, SIGTRAP, 0); 1036 } 1037 1038 userret(p, &frame, sticks); 1039 1040#ifdef KTRACE 1041 if (KTRPOINT(p, KTR_SYSRET)) 1042 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1043#endif 1044 1045 /* 1046 * This works because errno is findable through the 1047 * register set. If we ever support an emulation where this 1048 * is not the case, this code will need to be revisited. 1049 */ 1050 STOPEVENT(p, S_SCX, code); 1051 1052} 1053 1054/* 1055 * Simplified back end of syscall(), used when returning from fork() 1056 * directly into user mode. 1057 */ 1058void 1059fork_return(p, frame) 1060 struct proc *p; 1061 struct trapframe frame; 1062{ 1063 frame.tf_eax = 0; /* Child returns zero */ 1064 frame.tf_eflags &= ~PSL_C; /* success */ 1065 frame.tf_edx = 1; 1066 1067 userret(p, &frame, 0); 1068#ifdef KTRACE 1069 if (KTRPOINT(p, KTR_SYSRET)) 1070 ktrsysret(p->p_tracep, SYS_fork, 0, 0); 1071#endif 1072} 1073