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 * $FreeBSD$ 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/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/uio.h> 61 #include <sys/vmmeter.h> 62 #ifdef KTRACE 63 #include <sys/ktrace.h> 64 #endif 65 66 #include <vm/vm.h> 67 #include <vm/vm_param.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 #include <machine/vm86.h> 92 93 #ifdef DDB 94 extern int in_Debugger, debugger_on_panic; 95 #endif 96 97 #include "isa.h" 98 #include "npx.h" 99 100 int (*pmath_emulate) __P((struct trapframe *)); 101 102 extern void trap __P((struct trapframe frame)); 103 extern int trapwrite __P((unsigned addr)); 104 extern void syscall __P((struct trapframe frame)); 105 106 static int trap_pfault __P((struct trapframe *, int, vm_offset_t)); 107 static void trap_fatal __P((struct trapframe *, vm_offset_t)); 108 void dblfault_handler __P((void)); 109 110 extern inthand_t IDTVEC(syscall); 111 112 #define MAX_TRAP_MSG 28 113 static 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 }; 144 145 static __inline void userret __P((struct proc *p, struct trapframe *frame, 146 u_quad_t oticks)); 147 148 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 149 extern int has_f00f_bug; 150 #endif 151 152 static __inline void 153 userret(p, frame, oticks) 154 struct proc *p; 155 struct trapframe *frame; 156 u_quad_t oticks; 157 { 158 int sig, s; 159 160 while ((sig = CURSIG(p)) != 0) 161 postsig(sig); 162 163 #if 0 164 if (!want_resched && 165 (p->p_priority <= p->p_usrpri) && 166 (p->p_rtprio.type == RTP_PRIO_NORMAL)) { 167 int newpriority; 168 p->p_estcpu += 1; 169 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice; 170 newpriority = min(newpriority, MAXPRI); 171 p->p_usrpri = newpriority; 172 } 173 #endif 174 175 p->p_priority = p->p_usrpri; 176 if (want_resched) { 177 /* 178 * Since we are curproc, clock will normally just change 179 * our priority without moving us from one queue to another 180 * (since the running process is not on a queue.) 181 * If that happened after we setrunqueue ourselves but before we 182 * mi_switch()'ed, we might not be on the queue indicated by 183 * our priority. 184 */ 185 s = splhigh(); 186 setrunqueue(p); 187 p->p_stats->p_ru.ru_nivcsw++; 188 mi_switch(); 189 splx(s); 190 while ((sig = CURSIG(p)) != 0) 191 postsig(sig); 192 } 193 /* 194 * Charge system time if profiling. 195 */ 196 if (p->p_flag & P_PROFIL) 197 addupc_task(p, frame->tf_eip, 198 (u_int)(p->p_sticks - oticks) * psratio); 199 200 curpriority = p->p_priority; 201 } 202 203 /* 204 * Exception, fault, and trap interface to the FreeBSD kernel. 205 * This common code is called from assembly language IDT gate entry 206 * routines that prepare a suitable stack frame, and restore this 207 * frame after the exception has been processed. 208 */ 209 210 void 211 trap(frame) 212 struct trapframe frame; 213 { 214 struct proc *p = curproc; 215 u_quad_t sticks = 0; 216 int i = 0, ucode = 0, type, code; 217 vm_offset_t eva; 218 219 if (!(frame.tf_eflags & PSL_I)) { 220 /* 221 * Buggy application or kernel code has disabled interrupts 222 * and then trapped. Enabling interrupts now is wrong, but 223 * it is better than running with interrupts disabled until 224 * they are accidentally enabled later. 225 */ 226 type = frame.tf_trapno; 227 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) 228 printf( 229 "pid %ld (%s): trap %d with interrupts disabled\n", 230 (long)curproc->p_pid, curproc->p_comm, type); 231 else if (type != T_BPTFLT && type != T_TRCTRAP) 232 /* 233 * XXX not quite right, since this may be for a 234 * multiple fault in user mode. 235 */ 236 printf("kernel trap %d with interrupts disabled\n", 237 type); 238 enable_intr(); 239 } 240 241 eva = 0; 242 if (frame.tf_trapno == T_PAGEFLT) { 243 /* 244 * For some Cyrix CPUs, %cr2 is clobbered by interrupts. 245 * This problem is worked around by using an interrupt 246 * gate for the pagefault handler. We are finally ready 247 * to read %cr2 and then must reenable interrupts. 248 * 249 * XXX this should be in the switch statement, but the 250 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the 251 * flow of control too much for this to be obviously 252 * correct. 253 */ 254 eva = rcr2(); 255 enable_intr(); 256 } 257 258 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 259 restart: 260 #endif 261 type = frame.tf_trapno; 262 code = frame.tf_err; 263 264 if (in_vm86call) { 265 if (frame.tf_eflags & PSL_VM && 266 (type == T_PROTFLT || type == T_STKFLT)) { 267 i = vm86_emulate((struct vm86frame *)&frame); 268 if (i != 0) 269 /* 270 * returns to original process 271 */ 272 vm86_trap((struct vm86frame *)&frame); 273 return; 274 } 275 switch (type) { 276 /* 277 * these traps want either a process context, or 278 * assume a normal userspace trap. 279 */ 280 case T_PROTFLT: 281 case T_SEGNPFLT: 282 trap_fatal(&frame, eva); 283 return; 284 case T_TRCTRAP: 285 type = T_BPTFLT; /* kernel breakpoint */ 286 /* FALL THROUGH */ 287 } 288 goto kernel_trap; /* normal kernel trap handling */ 289 } 290 291 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) { 292 /* user trap */ 293 294 sticks = p->p_sticks; 295 p->p_md.md_regs = &frame; 296 297 switch (type) { 298 case T_PRIVINFLT: /* privileged instruction fault */ 299 ucode = type; 300 i = SIGILL; 301 break; 302 303 case T_BPTFLT: /* bpt instruction fault */ 304 case T_TRCTRAP: /* trace trap */ 305 frame.tf_eflags &= ~PSL_T; 306 i = SIGTRAP; 307 break; 308 309 case T_ARITHTRAP: /* arithmetic trap */ 310 ucode = code; 311 i = SIGFPE; 312 break; 313 314 case T_ASTFLT: /* Allow process switch */ 315 astoff(); 316 cnt.v_soft++; 317 if (p->p_flag & P_OWEUPC) { 318 p->p_flag &= ~P_OWEUPC; 319 addupc_task(p, p->p_stats->p_prof.pr_addr, 320 p->p_stats->p_prof.pr_ticks); 321 } 322 goto out; 323 324 /* 325 * The following two traps can happen in 326 * vm86 mode, and, if so, we want to handle 327 * them specially. 328 */ 329 case T_PROTFLT: /* general protection fault */ 330 case T_STKFLT: /* stack fault */ 331 if (frame.tf_eflags & PSL_VM) { 332 i = vm86_emulate((struct vm86frame *)&frame); 333 if (i == 0) 334 goto out; 335 break; 336 } 337 /* FALL THROUGH */ 338 339 case T_SEGNPFLT: /* segment not present fault */ 340 case T_TSSFLT: /* invalid TSS fault */ 341 case T_DOUBLEFLT: /* double fault */ 342 default: 343 ucode = code + BUS_SEGM_FAULT ; 344 i = SIGBUS; 345 break; 346 347 case T_PAGEFLT: /* page fault */ 348 i = trap_pfault(&frame, TRUE, eva); 349 if (i == -1) 350 return; 351 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 352 if (i == -2) 353 goto restart; 354 #endif 355 if (i == 0) 356 goto out; 357 358 ucode = T_PAGEFLT; 359 break; 360 361 case T_DIVIDE: /* integer divide fault */ 362 ucode = FPE_INTDIV; 363 i = SIGFPE; 364 break; 365 366 #if NISA > 0 367 case T_NMI: 368 #ifdef POWERFAIL_NMI 369 goto handle_powerfail; 370 #else /* !POWERFAIL_NMI */ 371 #ifdef DDB 372 /* NMI can be hooked up to a pushbutton for debugging */ 373 printf ("NMI ... going to debugger\n"); 374 if (kdb_trap (type, 0, &frame)) 375 return; 376 #endif /* DDB */ 377 /* machine/parity/power fail/"kitchen sink" faults */ 378 if (isa_nmi(code) == 0) return; 379 panic("NMI indicates hardware failure"); 380 #endif /* POWERFAIL_NMI */ 381 #endif /* NISA > 0 */ 382 383 case T_OFLOW: /* integer overflow fault */ 384 ucode = FPE_INTOVF; 385 i = SIGFPE; 386 break; 387 388 case T_BOUND: /* bounds check fault */ 389 ucode = FPE_FLTSUB; 390 i = SIGFPE; 391 break; 392 393 case T_DNA: 394 #if NNPX > 0 395 /* if a transparent fault (due to context switch "late") */ 396 if (npxdna()) 397 return; 398 #endif 399 if (!pmath_emulate) { 400 i = SIGFPE; 401 ucode = FPE_FPU_NP_TRAP; 402 break; 403 } 404 i = (*pmath_emulate)(&frame); 405 if (i == 0) { 406 if (!(frame.tf_eflags & PSL_T)) 407 return; 408 frame.tf_eflags &= ~PSL_T; 409 i = SIGTRAP; 410 } 411 /* else ucode = emulator_only_knows() XXX */ 412 break; 413 414 case T_FPOPFLT: /* FPU operand fetch fault */ 415 ucode = T_FPOPFLT; 416 i = SIGILL; 417 break; 418 } 419 } else { 420 kernel_trap: 421 /* kernel trap */ 422 423 switch (type) { 424 case T_PAGEFLT: /* page fault */ 425 (void) trap_pfault(&frame, FALSE, eva); 426 return; 427 428 case T_DNA: 429 #if NNPX > 0 430 /* 431 * The kernel is apparently using npx for copying. 432 * XXX this should be fatal unless the kernel has 433 * registered such use. 434 */ 435 if (npxdna()) 436 return; 437 #endif 438 break; 439 440 case T_PROTFLT: /* general protection fault */ 441 case T_SEGNPFLT: /* segment not present fault */ 442 /* 443 * Invalid segment selectors and out of bounds 444 * %eip's and %esp's can be set up in user mode. 445 * This causes a fault in kernel mode when the 446 * kernel tries to return to user mode. We want 447 * to get this fault so that we can fix the 448 * problem here and not have to check all the 449 * selectors and pointers when the user changes 450 * them. 451 */ 452 #define MAYBE_DORETI_FAULT(where, whereto) \ 453 do { \ 454 if (frame.tf_eip == (int)where) { \ 455 frame.tf_eip = (int)whereto; \ 456 return; \ 457 } \ 458 } while (0) 459 460 if (intr_nesting_level == 0) { 461 /* 462 * Invalid %fs's and %gs's can be created using 463 * procfs or PT_SETREGS or by invalidating the 464 * underlying LDT entry. This causes a fault 465 * in kernel mode when the kernel attempts to 466 * switch contexts. Lose the bad context 467 * (XXX) so that we can continue, and generate 468 * a signal. 469 */ 470 if (frame.tf_eip == (int)cpu_switch_load_gs) { 471 curpcb->pcb_gs = 0; 472 psignal(p, SIGBUS); 473 return; 474 } 475 MAYBE_DORETI_FAULT(doreti_iret, 476 doreti_iret_fault); 477 MAYBE_DORETI_FAULT(doreti_popl_ds, 478 doreti_popl_ds_fault); 479 MAYBE_DORETI_FAULT(doreti_popl_es, 480 doreti_popl_es_fault); 481 MAYBE_DORETI_FAULT(doreti_popl_fs, 482 doreti_popl_fs_fault); 483 if (curpcb && curpcb->pcb_onfault) { 484 frame.tf_eip = (int)curpcb->pcb_onfault; 485 return; 486 } 487 } 488 break; 489 490 case T_TSSFLT: 491 /* 492 * PSL_NT can be set in user mode and isn't cleared 493 * automatically when the kernel is entered. This 494 * causes a TSS fault when the kernel attempts to 495 * `iret' because the TSS link is uninitialized. We 496 * want to get this fault so that we can fix the 497 * problem here and not every time the kernel is 498 * entered. 499 */ 500 if (frame.tf_eflags & PSL_NT) { 501 frame.tf_eflags &= ~PSL_NT; 502 return; 503 } 504 break; 505 506 case T_TRCTRAP: /* trace trap */ 507 if (frame.tf_eip == (int)IDTVEC(syscall)) { 508 /* 509 * We've just entered system mode via the 510 * syscall lcall. Continue single stepping 511 * silently until the syscall handler has 512 * saved the flags. 513 */ 514 return; 515 } 516 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { 517 /* 518 * The syscall handler has now saved the 519 * flags. Stop single stepping it. 520 */ 521 frame.tf_eflags &= ~PSL_T; 522 return; 523 } 524 /* 525 * Fall through. 526 */ 527 case T_BPTFLT: 528 /* 529 * If DDB is enabled, let it handle the debugger trap. 530 * Otherwise, debugger traps "can't happen". 531 */ 532 #ifdef DDB 533 if (kdb_trap (type, 0, &frame)) 534 return; 535 #endif 536 break; 537 538 #if NISA > 0 539 case T_NMI: 540 #ifdef POWERFAIL_NMI 541 #ifndef TIMER_FREQ 542 # define TIMER_FREQ 1193182 543 #endif 544 handle_powerfail: 545 { 546 static unsigned lastalert = 0; 547 548 if(time_second - lastalert > 10) 549 { 550 log(LOG_WARNING, "NMI: power fail\n"); 551 sysbeep(TIMER_FREQ/880, hz); 552 lastalert = time_second; 553 } 554 return; 555 } 556 #else /* !POWERFAIL_NMI */ 557 #ifdef DDB 558 /* NMI can be hooked up to a pushbutton for debugging */ 559 printf ("NMI ... going to debugger\n"); 560 if (kdb_trap (type, 0, &frame)) 561 return; 562 #endif /* DDB */ 563 /* machine/parity/power fail/"kitchen sink" faults */ 564 if (isa_nmi(code) == 0) return; 565 /* FALL THROUGH */ 566 #endif /* POWERFAIL_NMI */ 567 #endif /* NISA > 0 */ 568 } 569 570 trap_fatal(&frame, eva); 571 return; 572 } 573 574 /* Translate fault for emulators (e.g. Linux) */ 575 if (*p->p_sysent->sv_transtrap) 576 i = (*p->p_sysent->sv_transtrap)(i, type); 577 578 trapsignal(p, i, ucode); 579 580 #ifdef DEBUG 581 if (type <= MAX_TRAP_MSG) { 582 uprintf("fatal process exception: %s", 583 trap_msg[type]); 584 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 585 uprintf(", fault VA = 0x%lx", (u_long)eva); 586 uprintf("\n"); 587 } 588 #endif 589 590 out: 591 userret(p, &frame, sticks); 592 } 593 594 #ifdef notyet 595 /* 596 * This version doesn't allow a page fault to user space while 597 * in the kernel. The rest of the kernel needs to be made "safe" 598 * before this can be used. I think the only things remaining 599 * to be made safe are the iBCS2 code and the process tracing/ 600 * debugging code. 601 */ 602 static int 603 trap_pfault(frame, usermode, eva) 604 struct trapframe *frame; 605 int usermode; 606 vm_offset_t eva; 607 { 608 vm_offset_t va; 609 struct vmspace *vm = NULL; 610 vm_map_t map = 0; 611 int rv = 0; 612 vm_prot_t ftype; 613 struct proc *p = curproc; 614 615 if (frame->tf_err & PGEX_W) 616 ftype = VM_PROT_READ | VM_PROT_WRITE; 617 else 618 ftype = VM_PROT_READ; 619 620 va = trunc_page(eva); 621 if (va < VM_MIN_KERNEL_ADDRESS) { 622 vm_offset_t v; 623 vm_page_t mpte; 624 625 if (p == NULL || 626 (!usermode && va < VM_MAXUSER_ADDRESS && 627 (intr_nesting_level != 0 || curpcb == NULL || 628 curpcb->pcb_onfault == NULL))) { 629 trap_fatal(frame, eva); 630 return (-1); 631 } 632 633 /* 634 * This is a fault on non-kernel virtual memory. 635 * vm is initialized above to NULL. If curproc is NULL 636 * or curproc->p_vmspace is NULL the fault is fatal. 637 */ 638 vm = p->p_vmspace; 639 if (vm == NULL) 640 goto nogo; 641 642 map = &vm->vm_map; 643 644 /* 645 * Keep swapout from messing with us during this 646 * critical time. 647 */ 648 ++p->p_lock; 649 650 /* 651 * Grow the stack if necessary 652 */ 653 /* grow_stack returns false only if va falls into 654 * a growable stack region and the stack growth 655 * fails. It returns true if va was not within 656 * a growable stack region, or if the stack 657 * growth succeeded. 658 */ 659 if (!grow_stack (p, va)) { 660 rv = KERN_FAILURE; 661 --p->p_lock; 662 goto nogo; 663 } 664 665 /* Fault in the user page: */ 666 rv = vm_fault(map, va, ftype, 667 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0); 668 669 --p->p_lock; 670 } else { 671 /* 672 * Don't allow user-mode faults in kernel address space. 673 */ 674 if (usermode) 675 goto nogo; 676 677 /* 678 * Since we know that kernel virtual address addresses 679 * always have pte pages mapped, we just have to fault 680 * the page. 681 */ 682 rv = vm_fault(kernel_map, va, ftype, FALSE); 683 } 684 685 if (rv == KERN_SUCCESS) 686 return (0); 687 nogo: 688 if (!usermode) { 689 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 690 frame->tf_eip = (int)curpcb->pcb_onfault; 691 return (0); 692 } 693 trap_fatal(frame, eva); 694 return (-1); 695 } 696 697 /* kludge to pass faulting virtual address to sendsig */ 698 frame->tf_err = eva; 699 700 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 701 } 702 #endif 703 704 int 705 trap_pfault(frame, usermode, eva) 706 struct trapframe *frame; 707 int usermode; 708 vm_offset_t eva; 709 { 710 vm_offset_t va; 711 struct vmspace *vm = NULL; 712 vm_map_t map = 0; 713 int rv = 0; 714 vm_prot_t ftype; 715 struct proc *p = curproc; 716 717 va = trunc_page(eva); 718 if (va >= KERNBASE) { 719 /* 720 * Don't allow user-mode faults in kernel address space. 721 * An exception: if the faulting address is the invalid 722 * instruction entry in the IDT, then the Intel Pentium 723 * F00F bug workaround was triggered, and we need to 724 * treat it is as an illegal instruction, and not a page 725 * fault. 726 */ 727 #if defined(I586_CPU) && !defined(NO_F00F_HACK) 728 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) { 729 frame->tf_trapno = T_PRIVINFLT; 730 return -2; 731 } 732 #endif 733 if (usermode) 734 goto nogo; 735 736 map = kernel_map; 737 } else { 738 /* 739 * This is a fault on non-kernel virtual memory. 740 * vm is initialized above to NULL. If curproc is NULL 741 * or curproc->p_vmspace is NULL the fault is fatal. 742 */ 743 if (p != NULL) 744 vm = p->p_vmspace; 745 746 if (vm == NULL) 747 goto nogo; 748 749 map = &vm->vm_map; 750 } 751 752 if (frame->tf_err & PGEX_W) 753 ftype = VM_PROT_READ | VM_PROT_WRITE; 754 else 755 ftype = VM_PROT_READ; 756 757 if (map != kernel_map) { 758 /* 759 * Keep swapout from messing with us during this 760 * critical time. 761 */ 762 ++p->p_lock; 763 764 /* 765 * Grow the stack if necessary 766 */ 767 /* grow_stack returns false only if va falls into 768 * a growable stack region and the stack growth 769 * fails. It returns true if va was not within 770 * a growable stack region, or if the stack 771 * growth succeeded. 772 */ 773 if (!grow_stack (p, va)) { 774 rv = KERN_FAILURE; 775 --p->p_lock; 776 goto nogo; 777 } 778 779 /* Fault in the user page: */ 780 rv = vm_fault(map, va, ftype, 781 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY : 0); 782 783 --p->p_lock; 784 } else { 785 /* 786 * Don't have to worry about process locking or stacks in the kernel. 787 */ 788 rv = vm_fault(map, va, ftype, FALSE); 789 } 790 791 if (rv == KERN_SUCCESS) 792 return (0); 793 nogo: 794 if (!usermode) { 795 if (intr_nesting_level == 0 && curpcb && curpcb->pcb_onfault) { 796 frame->tf_eip = (int)curpcb->pcb_onfault; 797 return (0); 798 } 799 trap_fatal(frame, eva); 800 return (-1); 801 } 802 803 /* kludge to pass faulting virtual address to sendsig */ 804 frame->tf_err = eva; 805 806 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 807 } 808 809 static void 810 trap_fatal(frame, eva) 811 struct trapframe *frame; 812 vm_offset_t eva; 813 { 814 int code, type, ss, esp; 815 struct soft_segment_descriptor softseg; 816 817 code = frame->tf_err; 818 type = frame->tf_trapno; 819 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); 820 821 if (type <= MAX_TRAP_MSG) 822 printf("\n\nFatal trap %d: %s while in %s mode\n", 823 type, trap_msg[type], 824 frame->tf_eflags & PSL_VM ? "vm86" : 825 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 826 #ifdef SMP 827 /* three seperate prints in case of a trap on an unmapped page */ 828 printf("mp_lock = %08x; ", mp_lock); 829 printf("cpuid = %d; ", cpuid); 830 printf("lapic.id = %08x\n", lapic.id); 831 #endif 832 if (type == T_PAGEFLT) { 833 printf("fault virtual address = 0x%x\n", eva); 834 printf("fault code = %s %s, %s\n", 835 code & PGEX_U ? "user" : "supervisor", 836 code & PGEX_W ? "write" : "read", 837 code & PGEX_P ? "protection violation" : "page not present"); 838 } 839 printf("instruction pointer = 0x%x:0x%x\n", 840 frame->tf_cs & 0xffff, frame->tf_eip); 841 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { 842 ss = frame->tf_ss & 0xffff; 843 esp = frame->tf_esp; 844 } else { 845 ss = GSEL(GDATA_SEL, SEL_KPL); 846 esp = (int)&frame->tf_esp; 847 } 848 printf("stack pointer = 0x%x:0x%x\n", ss, esp); 849 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); 850 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", 851 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 852 printf(" = DPL %d, pres %d, def32 %d, gran %d\n", 853 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, 854 softseg.ssd_gran); 855 printf("processor eflags = "); 856 if (frame->tf_eflags & PSL_T) 857 printf("trace trap, "); 858 if (frame->tf_eflags & PSL_I) 859 printf("interrupt enabled, "); 860 if (frame->tf_eflags & PSL_NT) 861 printf("nested task, "); 862 if (frame->tf_eflags & PSL_RF) 863 printf("resume, "); 864 if (frame->tf_eflags & PSL_VM) 865 printf("vm86, "); 866 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); 867 printf("current process = "); 868 if (curproc) { 869 printf("%lu (%s)\n", 870 (u_long)curproc->p_pid, curproc->p_comm ? 871 curproc->p_comm : ""); 872 } else { 873 printf("Idle\n"); 874 } 875 printf("interrupt mask = "); 876 if ((cpl & net_imask) == net_imask) 877 printf("net "); 878 if ((cpl & tty_imask) == tty_imask) 879 printf("tty "); 880 if ((cpl & bio_imask) == bio_imask) 881 printf("bio "); 882 if ((cpl & cam_imask) == cam_imask) 883 printf("cam "); 884 if (cpl == 0) 885 printf("none"); 886 #ifdef SMP 887 /** 888 * XXX FIXME: 889 * we probably SHOULD have stopped the other CPUs before now! 890 * another CPU COULD have been touching cpl at this moment... 891 */ 892 printf(" <- SMP: XXX"); 893 #endif 894 printf("\n"); 895 896 #ifdef KDB 897 if (kdb_trap(&psl)) 898 return; 899 #endif 900 #ifdef DDB 901 if ((debugger_on_panic || in_Debugger) && kdb_trap(type, 0, frame)) 902 return; 903 #endif 904 printf("trap number = %d\n", type); 905 if (type <= MAX_TRAP_MSG) 906 panic(trap_msg[type]); 907 else 908 panic("unknown/reserved trap"); 909 } 910 911 /* 912 * Double fault handler. Called when a fault occurs while writing 913 * a frame for a trap/exception onto the stack. This usually occurs 914 * when the stack overflows (such is the case with infinite recursion, 915 * for example). 916 * 917 * XXX Note that the current PTD gets replaced by IdlePTD when the 918 * task switch occurs. This means that the stack that was active at 919 * the time of the double fault is not available at <kstack> unless 920 * the machine was idle when the double fault occurred. The downside 921 * of this is that "trace <ebp>" in ddb won't work. 922 */ 923 void 924 dblfault_handler() 925 { 926 printf("\nFatal double fault:\n"); 927 printf("eip = 0x%x\n", common_tss.tss_eip); 928 printf("esp = 0x%x\n", common_tss.tss_esp); 929 printf("ebp = 0x%x\n", common_tss.tss_ebp); 930 #ifdef SMP 931 /* three seperate prints in case of a trap on an unmapped page */ 932 printf("mp_lock = %08x; ", mp_lock); 933 printf("cpuid = %d; ", cpuid); 934 printf("lapic.id = %08x\n", lapic.id); 935 #endif 936 panic("double fault"); 937 } 938 939 /* 940 * Compensate for 386 brain damage (missing URKR). 941 * This is a little simpler than the pagefault handler in trap() because 942 * it the page tables have already been faulted in and high addresses 943 * are thrown out early for other reasons. 944 */ 945 int trapwrite(addr) 946 unsigned addr; 947 { 948 struct proc *p; 949 vm_offset_t va; 950 struct vmspace *vm; 951 int rv; 952 953 va = trunc_page((vm_offset_t)addr); 954 /* 955 * XXX - MAX is END. Changed > to >= for temp. fix. 956 */ 957 if (va >= VM_MAXUSER_ADDRESS) 958 return (1); 959 960 p = curproc; 961 vm = p->p_vmspace; 962 963 ++p->p_lock; 964 965 if (!grow_stack (p, va)) { 966 --p->p_lock; 967 return (1); 968 } 969 970 /* 971 * fault the data page 972 */ 973 rv = vm_fault(&vm->vm_map, va, VM_PROT_READ|VM_PROT_WRITE, VM_FAULT_DIRTY); 974 975 --p->p_lock; 976 977 if (rv != KERN_SUCCESS) 978 return 1; 979 980 return (0); 981 } 982 983 /* 984 * System call request from POSIX system call gate interface to kernel. 985 * Like trap(), argument is call by reference. 986 */ 987 void 988 syscall(frame) 989 struct trapframe frame; 990 { 991 caddr_t params; 992 int i; 993 struct sysent *callp; 994 struct proc *p = curproc; 995 u_quad_t sticks; 996 int error; 997 int args[8]; 998 u_int code; 999 1000 #ifdef DIAGNOSTIC 1001 if (ISPL(frame.tf_cs) != SEL_UPL) 1002 panic("syscall"); 1003 #endif 1004 sticks = p->p_sticks; 1005 p->p_md.md_regs = &frame; 1006 params = (caddr_t)frame.tf_esp + sizeof(int); 1007 code = frame.tf_eax; 1008 if (p->p_sysent->sv_prepsyscall) { 1009 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); 1010 } else { 1011 /* 1012 * Need to check if this is a 32 bit or 64 bit syscall. 1013 */ 1014 if (code == SYS_syscall) { 1015 /* 1016 * Code is first argument, followed by actual args. 1017 */ 1018 code = fuword(params); 1019 params += sizeof(int); 1020 } else if (code == SYS___syscall) { 1021 /* 1022 * Like syscall, but code is a quad, so as to maintain 1023 * quad alignment for the rest of the arguments. 1024 */ 1025 code = fuword(params); 1026 params += sizeof(quad_t); 1027 } 1028 } 1029 1030 if (p->p_sysent->sv_mask) 1031 code &= p->p_sysent->sv_mask; 1032 1033 if (code >= p->p_sysent->sv_size) 1034 callp = &p->p_sysent->sv_table[0]; 1035 else 1036 callp = &p->p_sysent->sv_table[code]; 1037 1038 if (params && (i = callp->sy_narg * sizeof(int)) && 1039 (error = copyin(params, (caddr_t)args, (u_int)i))) { 1040 #ifdef KTRACE 1041 if (KTRPOINT(p, KTR_SYSCALL)) 1042 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 1043 #endif 1044 goto bad; 1045 } 1046 #ifdef KTRACE 1047 if (KTRPOINT(p, KTR_SYSCALL)) 1048 ktrsyscall(p->p_tracep, code, callp->sy_narg, args); 1049 #endif 1050 p->p_retval[0] = 0; 1051 p->p_retval[1] = frame.tf_edx; 1052 1053 STOPEVENT(p, S_SCE, callp->sy_narg); 1054 1055 error = (*callp->sy_call)(p, args); 1056 1057 switch (error) { 1058 1059 case 0: 1060 /* 1061 * Reinitialize proc pointer `p' as it may be different 1062 * if this is a child returning from fork syscall. 1063 */ 1064 p = curproc; 1065 frame.tf_eax = p->p_retval[0]; 1066 frame.tf_edx = p->p_retval[1]; 1067 frame.tf_eflags &= ~PSL_C; 1068 break; 1069 1070 case ERESTART: 1071 /* 1072 * Reconstruct pc, assuming lcall $X,y is 7 bytes, 1073 * int 0x80 is 2 bytes. We saved this in tf_err. 1074 */ 1075 frame.tf_eip -= frame.tf_err; 1076 break; 1077 1078 case EJUSTRETURN: 1079 break; 1080 1081 default: 1082 bad: 1083 if (p->p_sysent->sv_errsize) { 1084 if (error >= p->p_sysent->sv_errsize) 1085 error = -1; /* XXX */ 1086 else 1087 error = p->p_sysent->sv_errtbl[error]; 1088 } 1089 frame.tf_eax = error; 1090 frame.tf_eflags |= PSL_C; 1091 break; 1092 } 1093 1094 if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) { 1095 /* Traced syscall. */ 1096 frame.tf_eflags &= ~PSL_T; 1097 trapsignal(p, SIGTRAP, 0); 1098 } 1099 1100 userret(p, &frame, sticks); 1101 1102 #ifdef KTRACE 1103 if (KTRPOINT(p, KTR_SYSRET)) 1104 ktrsysret(p->p_tracep, code, error, p->p_retval[0]); 1105 #endif 1106 1107 /* 1108 * This works because errno is findable through the 1109 * register set. If we ever support an emulation where this 1110 * is not the case, this code will need to be revisited. 1111 */ 1112 STOPEVENT(p, S_SCX, code); 1113 1114 } 1115 1116 /* 1117 * Simplified back end of syscall(), used when returning from fork() 1118 * directly into user mode. 1119 */ 1120 void 1121 fork_return(p, frame) 1122 struct proc *p; 1123 struct trapframe frame; 1124 { 1125 frame.tf_eax = 0; /* Child returns zero */ 1126 frame.tf_eflags &= ~PSL_C; /* success */ 1127 frame.tf_edx = 1; 1128 1129 userret(p, &frame, 0); 1130 #ifdef KTRACE 1131 if (KTRPOINT(p, KTR_SYSRET)) 1132 ktrsysret(p->p_tracep, SYS_fork, 0, 0); 1133 #endif 1134 } 1135